A new method for generating low-probability risk scenarios of a power distribution network with host access

Abstract

The application discloses a novel main body access power distribution network low-probability risk scene generation method, and the method comprises the following steps: based on the historical operation data of a target power distribution network, determining source and load operation states and identifying low-density operation samples, and determining the low-density operation samples meeting preset data validity conditions and preset risk discrimination conditions as risk operation samples; determining joint operation states according to the source and load operation states and the marking results of the risk operation samples, forming a historical joint operation state sequence and constructing a Markov state transition model; generating a target time period joint operation state sequence based on the model, and generating a power distribution network low-probability risk scene according to the joint operation state sequence and the corresponding operation samples. The application can retain low-frequency risk operation characteristics, reduce the possibility that risk samples are ignored in the scene generation process, and improve the pertinence of low-probability risk scene generation.

Description

Technical Field

[0001] This invention relates to the field of distribution network operation analysis and scenario generation technology, specifically to a novel method for generating low-probability risk scenarios in distribution networks with active participants. Background Technology

[0002] With the continuous integration of new entities such as distributed power sources, energy storage devices, electric vehicles, and adjustable loads into distribution networks, the source-load relationships, power flow, and operational risks of distribution networks exhibit stronger randomness and temporal variation characteristics. To assess the operational safety of distribution networks under different integration conditions, source-load operational characteristics are typically extracted based on historical operational data, and distribution network operation scenarios are generated using methods such as cluster analysis, state transition modeling, and sample reconstruction.

[0003] Existing scenario generation methods often discard low-density operational samples as outliers or noise. However, low-density operational samples may be caused by missing data or measurement anomalies, or they may correspond to real low-frequency risk conditions such as voltage overruns, equipment overloads, or reverse power flows. Without further differentiation of their data validity and risk attributes, risky operational samples with safety analysis value may be mistakenly deleted.

[0004] Meanwhile, existing methods mostly construct state transition relationships based on conventional source-load operation states, without incorporating the risk markers of low-density risk operation samples into the operation state and its temporal transition process. This results in the generated target time period operation state sequence mainly reflecting high-frequency conventional operating conditions, making it difficult to effectively reproduce low-frequency risk states and their evolution process. Consequently, the generated scenario does not adequately cover low-probability risk operating conditions under the access conditions of new subjects.

[0005] Therefore, there is an urgent need for a novel method for generating subject-based distribution network scenarios that can identify and retain low-density risk operation samples, incorporate risk operation characteristics into state transition modeling, and generate low-probability risk scenarios. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art, the objective of this invention is to provide a novel method for generating low-probability risk scenarios in a distribution network with primary access, comprising: Based on the historical operation data of the target distribution network, determine the source and load operation status corresponding to the historical operation samples, and identify low-density operation samples; The low-density operation samples that meet the preset data validity conditions and preset risk discrimination conditions are identified as risk operation samples; Based on the source load operating status and the labeling results of the risk operating samples, the joint operating status is determined, a historical joint operating status sequence is formed according to the sampling time, and a Markov state transition model is constructed based on the historical joint operating status sequence. Based on the Markov state transition model, a joint operating state sequence for the target time period is generated, and based on the joint operating state sequence and the corresponding operating samples, a low-probability risk scenario for the distribution network containing risk operation characteristics is generated.

[0007] This invention uses historical operating samples of the target distribution network as the processing object. On the one hand, it determines the operating status of the source and load based on the source and load operating characteristics corresponding to the historical operating samples. On the other hand, it identifies operating samples with low distribution density in the sample space from the historical operating samples. Since low-density operating samples may be caused by data problems such as missing data or measurement anomalies, or they may correspond to low-frequency risk conditions that actually occurred during historical operation, the availability of data in low-density operating samples is judged by preset data validity conditions, and samples with risky operating characteristics are identified by preset risk discrimination conditions, thus obtaining risky operating samples.

[0008] Based on this, the labeling results of source load operating states and risk operating samples are combined so that the joint operating state simultaneously includes the source load operating conditions and risk attributes corresponding to the historical operating samples. The joint operating states corresponding to each historical sampling time are arranged according to the sampling time to form a historical joint operating state sequence. A Markov state transition model is constructed based on the transition relationship between adjacent joint operating states to characterize the temporal evolution relationship between various joint operating states, including risk operating states.

[0009] Based on the Markov state transition model, a joint operating state sequence for the target period is generated according to the state transition relationship corresponding to each sampling time within the target period. Then, based on each joint operating state in the joint operating state sequence, an operating sample with corresponding source-load operating conditions and risk attributes is called to convert the discrete joint operating states into specific combinations of operating variables, thereby forming a low-probability risk scenario for the distribution network that includes risk operation characteristics.

[0010] Compared with the prior art, the embodiments of the present invention have at least the following beneficial effects: 1. This invention sequentially performs data validity and risk assessment on low-density operating samples, distinguishing invalid low-density samples formed due to data loss, abnormal collection, etc., from low-density operating samples with actual safety risks, and identifying low-density operating samples that meet the corresponding assessment conditions as risky operating samples. This can preserve low-frequency risk information in historical operating data and reduce the possibility of low-probability risk conditions being eliminated as noise during density clustering.

[0011] 2. This invention combines the source-load operating status with the labeling results of risk operating samples to form a joint operating status, and constructs a Markov state transition model based on the historical joint operating status sequence, so that risk operating attributes participate in the time-series transition modeling of distribution network operating status; on this basis, a joint operating status sequence for the target time period is generated based on the Markov state transition model, and the corresponding operating samples are called to generate low-probability risk scenarios of the distribution network, so that the generated scenarios can characterize low-probability risk conditions and their state evolution process, and improve the coverage of low-probability risk scenarios under the condition of new subject access. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0013] Figure 1 A schematic diagram of the overall process for generating low-probability risk scenarios in a distribution network for a new type of entity access; Figure 2 This is a schematic diagram illustrating the specific implementation process of a method for generating low-probability risk scenarios in a distribution network with a novel type of entity access. Detailed Implementation

[0014] The technical solution of the present invention will be clearly and completely described below with reference to embodiments thereof. Obviously, the described embodiments are merely one specific implementation of the present invention, and not all implementations. Other implementations obtained by those skilled in the art based on the embodiments of the present invention without creative effort should all fall within the protection scope of the present invention.

[0015] This embodiment provides a method for generating low-probability risk scenarios in a distribution network with novel entities connected. The novel entities include one or more of distributed photovoltaic (PV), distributed wind power, energy storage devices, electric vehicle charging facilities, and adjustable loads.

[0016] First, historical operational data for the target distribution network over a continuous historical period is acquired. This historical operational data includes distribution network operational data, new entity operational data, and data related to security access risks. Distribution network operational data includes node voltage, feeder active power, feeder reactive power, node active load, node reactive load, line load rate, and distribution transformer load rate. New entity operational data includes distributed generation active power output, energy storage charging and discharging power, electric vehicle charging and discharging power, and adjustable load response power. Security access risk-related data includes data used to identify voltage exceedances, line overloads, distribution transformer overloads, reverse power flow, and net load ramp-up exceedances. For energy storage devices and electric vehicle charging facilities, their discharge power is defined as positive, and their charging power as negative, to unify the active power direction of different new entities. Energy storage devices are not directly used for calculating source-load correlation coefficients; instead, they participate in subsequent scenario reconstruction and operational constraint verification through energy storage charging and discharging power and state of charge.

[0017] Then, time alignment is performed on different data sources according to a unified sampling period. For a single missing data point, linear interpolation is performed using the effective measurement values ​​adjacent to the missing location to complete the missing data. For measurement values ​​exceeding the effective range of the measurement device or the physical value range of the equipment, they are marked as abnormal measurement values ​​and corrected using adjacent effective measurement values. For abnormal measurement data with continuous missing data or that cannot be corrected, the corresponding abnormal markers are retained for subsequent data validity assessment. After the above data preprocessing, only historical operating samples where all operating variables have usable values ​​and can form a complete joint feature vector are included in subsequent standardization processing, source-load operating state division, and density clustering. Historical operating samples that still have continuous missing data or cannot obtain usable values ​​and cannot form a complete joint feature vector are not included in subsequent state division, density clustering, and state transition modeling, and the corresponding abnormal records are retained. For historical operating samples that can form a complete joint feature vector after interpolation repair, their abnormal markers and repair records are retained for step S201 to further assess the data validity of historical operating samples identified as low-density operating samples.

[0018] Subsequently, the continuous variables of different dimensions involved in the sample distance calculation were standardized. Sampling time The corresponding number The standardized value of a class runtime variable is represented as:

[0019] In the formula, Indicates the sampling time index of historical running data; Indicates the sampling time The corresponding number The original measurement value of the class's runtime variable; This represents the corresponding standardized value; and They represent the first The mean and standard deviation of the class's runtime variables over a preset historical period; Indicates the category index of the variable being run. When the... Class runtime variables remain unchanged within a preset historical period, thus When this happens, the corresponding standardized value is set to 0 to avoid the denominator being zero in the standardized calculation.

[0020] Standardization is used to eliminate differences in the dimensions and numerical ranges of operating variables such as power, voltage, and equipment load rate. The standardized values ​​are used only in step S103 to calculate the feature distance between historical operating samples, and in step S406 to calculate the operating feature distance and temporal morphological distance between candidate scenarios; they are not used to construct new joint feature vectors or sample sets independently. Source-load correlation calculation, source-load operating state classification, risk identification, scenario reconstruction, and operating constraint verification all use the original physical quantities of the corresponding operating variables.

[0021] The preprocessed operational data are combined according to the sampling time to construct a joint feature set. Sampling time Corresponding joint feature vector Represented as:

[0022] In the formula, This represents the active power characteristic vector of distributed power sources and adjustable resources. This represents the active power load characteristic vector; Represents the reactive load characteristic vector; Represents the node voltage eigenvector; This represents the characteristic vector of the load rate of the line and distribution transformer. This represents a feature vector related to secure access risks, where the joint feature vector... Constituting sampling time The corresponding historical samples are composed of joint feature vectors corresponding to all historical sampling times, forming a joint feature set:

[0023] In the formula, This represents the joint feature set, which is also the set of historical operation samples; This represents the total number of historical running samples.

[0024] It should be noted that the feature vectors related to secure access risks include reverse power flow, net load ramp rate, and other features not included in the node voltage feature vectors. and load factor characteristic vectors of power lines and distribution transformers The derived risk characteristics are repeatedly represented in the text.

[0025] like Figure 1 and Figure 2 As shown, this embodiment provides a novel method for generating low-probability risk scenarios in a distribution network with primary access, along with a specific implementation process, which includes the following steps: S1. Based on the historical operation data of the target distribution network, determine the source and load operation status corresponding to the historical operation sample and identify low-density operation samples; This step first uses distributed power output data and load power data in the joint feature set to extract source-load correlation, and then combines active load level, renewable energy output level and source-load correlation to classify source-load operating status; subsequently, density clustering is performed on the joint feature set to identify low-density operating samples in the joint feature set that are not classified into the regular operating sample cluster.

[0026] In this embodiment, step S1 includes the following steps: S101. Based on the distributed power output data and load power data in the historical operation samples, extract the source-load correlation to characterize the source-load operation relationship; Both distributed generation output and load power change continuously over time, and the power value at a single sampling moment cannot fully represent the coordinated change relationship between the two. Therefore, this embodiment uses each historical sampling moment as the endpoint to extract the distributed generation output sequence and load power sequence over a continuous time range, and calculates the source-load correlation accordingly.

[0027] S1011. Within a preset time window, calculate the source-load correlation coefficient between the output sequence of each distributed power source access point and the load power sequence of each load area. With the first Taking the sampling time as the endpoint, select consecutive times forward... A number of sampling points are used to form a preset correlation calculation time period:

[0028] In the formula, Indicates the sampling time The corresponding historical operating samples and the relevant time period used for correlation calculation; This indicates the number of consecutive sampling points included in the correlation calculation period. The period is determined based on the sampling period of historical data and the period of source load change that needs to be reflected.

[0029] In this embodiment, when the historical data sampling period is 15 minutes and it is necessary to reflect the intraday source-load variation relationship, the following can be used: Setting it to 96 ensures that the correlation calculation period covers a continuous 24 hours. Using a continuous 24-hour period for correlation calculation can simultaneously cover low-load periods at night, periods of load variation in the morning and evening, and periods of variable output of distributed photovoltaic power during the day, reducing the impact of short-term, accidental fluctuations on the correlation calculation results when only a partial time period is used.

[0030] For the The first distributed power access point and the first Each load region is derived from the joint feature set. Extract the active power output sequence and active power load sequence within the correlation calculation period, and calculate the source-load correlation coefficient:

[0031] In the formula, Indicates the first The distributed power access point and the first The source-load correlation coefficient of each load area within the correlation calculation period; Indicates the first At the sampling time, each distributed power access point Those who have made meritorious contributions; Indicates the first Each load region at the sampling time Active load; and These represent the standard deviations of the corresponding active power output sequence and active power load sequence during the correlation calculation period, respectively. Indicates the period for correlation calculation Index of historical sampling times within.

[0032] The range of values ​​for the source-load correlation coefficient is as follows: When the source-load correlation coefficient is greater than zero, it indicates that the output of distributed power sources and the load power change in the same direction as a whole; when the source-load correlation coefficient is less than zero, it indicates that the two change in opposite directions as a whole; when the source-load correlation coefficient is close to zero, it indicates that the two do not show a significant linear correlation relationship within the current correlation calculation period.

[0033] When the output sequence of distributed power sources or the power sequence of loads remains unchanged during the correlation calculation period, resulting in a corresponding standard deviation of zero, the corresponding source-load correlation coefficient is set to zero to indicate that no identifiable synchronous change relationship has been formed during the current correlation calculation period.

[0034] In this step and All of these correspond to historical operational data, among which, This represents the index of the last historical sampling time for the correlation calculation period. This represents the index of historical sampling times within the correlation calculation period; the target sampling time within the subsequent target period is represented by... express.

[0035] S1012. Arrange the source-load correlation coefficients according to the correspondence between the distributed power supply access point and the load area, construct the source-load correlation matrix, and determine the source-load correlation based on the source-load correlation matrix. Since the power variation relationships between different distributed power source access points and different load areas differ, calculating a single correlation coefficient based solely on the total active power output and total active power load of the target distribution network can easily obscure the source-load coordinated variation characteristics within local areas. Therefore, this embodiment retains the spatial correspondence between each distributed power source access point and each load area, and arranges the source-load correlation coefficients obtained in step S1011 accordingly.

[0036] Specifically, the matrix rows are arranged according to the predetermined distributed power supply access point numbers, and the matrix columns are arranged according to the predetermined load area numbers, ensuring that the same row and column arrangement order is used at different sampling times. Among these, the matrix... The row corresponds to the first The first distributed power access point, the... Column corresponding to the first The load area, the first Line number The matrix elements of the column are the two in the th column. The correlation coefficients of the source and load within the time period corresponding to each sampling time are calculated, thereby constructing the source and load correlation matrix:

[0037] In the formula, Indicates the sampling time The corresponding source-load correlation matrix; Indicates the number of distributed power supply access points; Indicates the number of load areas.

[0038] Through the above arrangement, each element in the source-load correlation matrix has a definite spatial correspondence, and the source-load correlation matrices obtained at different times can be compared element-by-element according to the same access point and load area. Compared to calculating only a single correlation coefficient between the total active power output and total active power load of the entire network, the source-load correlation matrix can preserve the local source-load correlation differences between different access points and different load areas. The source-load correlation matrix includes... The source-load correlation coefficient can comprehensively characterize the local source-load correlation relationships of the target distribution network. However, the subsequent source-load operating status needs to be classified into levels by combining the active power load level, the renewable energy output level, and the source-load correlation. If each matrix element in the source-load correlation matrix is ​​directly classified into levels, it will increase the combination dimension of the source-load operating status and make the number of historical samples corresponding to some joint states too small, affecting the stable statistics of subsequent state transition probabilities.

[0039] Therefore, by summing all the source-load correlation coefficients in the source-load correlation matrix, a source-load correlation index for classifying source-load operating states is obtained:

[0040] In the formula, Indicates the sampling time The corresponding source-load correlation index retains the positive and negative directions of each source-load correlation coefficient, and is used to characterize the average source-load correlation direction and average correlation degree of the target distribution network within the current correlation calculation period. Among them, when... When the value is positive, it indicates that the output of distributed generation and the load power in the target distribution network generally tend to change in the same direction; when... When the value is negative, it indicates that the two tend to change in opposite directions overall; when When the value is close to zero, it indicates that there is mutual cancellation between different local source-load correlations, or that no obvious source-load linear correlation has been formed in the target distribution network as a whole.

[0041] It should be noted that the source-load correlation index is used to transform source-load correlation into a scalar feature that can be classified together with active power load level index and new energy output level index, rather than to replace the complete source-load correlation matrix.

[0042] Therefore, the source-load correlation index With sampling time A correspondence is established between historical operation samples, and this is input into step S102 to determine the source-load correlation level corresponding to the historical operation sample; at the same time, the complete source-load correlation matrix is ​​retained. This is used for subsequent source-load correlation maintenance verification of candidate scenarios and for calculating the source-load correlation distance between candidate scenarios.

[0043] S102. Combining the source-load operation characteristics and source-load correlation of historical operation samples, the historical operation samples are divided into states to obtain the source-load operation states. The source-load correlation index obtained in step S101 is used to characterize the average change relationship between the output of distributed power sources and the load power during the preset correlation calculation period. However, the source-load correlation alone cannot fully distinguish the source-load operating conditions of the power grid.

[0044] Source-load correlation can characterize the coordinated change relationship between distributed generation output and load power, but the same source-load correlation may correspond to different power levels. For example, a positive correlation may occur under low load and low renewable energy output conditions, or under high load and high renewable energy output conditions; similarly, the same high renewable energy output level will have different impacts on reverse power flow, node voltage, and equipment load rate under different load levels. Therefore, this embodiment further extracts active power load level and renewable energy output level from historical operating samples, and uses both and source-load correlation together for source-load operating state classification.

[0045] First, summarize the sampling times. The active load of each corresponding load area is used to obtain the total active load of the target distribution network. Simultaneously, the active power output of each distributed photovoltaic and distributed wind power grid connection point is aggregated to obtain the total active power output of new energy sources. .

[0046] Based on the total active load and the total active power output of new energy sources, calculate the active load level index and the new energy output level index respectively:

[0047]

[0048] In the formula, Indicates the sampling time The corresponding active power load level index; This indicates the maximum active power load within the preset historical statistical period; Indicates the sampling time Corresponding new energy output level indicators; This represents the total installed capacity of distributed photovoltaic and distributed wind power within the target distribution network. The active load level is determined based on historical load data after the missing measurements have been completed and the abnormal measurements have been corrected, in order to avoid the overall active load level index being too low due to abnormal measurement peaks. The renewable energy equipment is determined based on the historical data of the target distribution network for the corresponding period. When the installed capacity of renewable energy changes within the historical statistical period, the renewable energy output level index is calculated according to the actual installed capacity at each sampling time.

[0049] The active load level index reflects the degree of occupancy of current load demand relative to the historical maximum load; the renewable energy output level index reflects the degree of utilization of current renewable energy output relative to installed capacity; and the source-load correlation index reflects the average coordinated change direction of renewable energy output and load power. These three types of indicators describe the operating conditions of the distribution network from three aspects: load demand scale, renewable energy supply scale, and source-load coordination relationship.

[0050] After obtaining the three types of indicators, statistics were compiled separately. , and In joint feature set The historical distribution of data was analyzed, and the grading boundaries were determined according to the same statistical caliber.

[0051] This embodiment uses an equal-frequency partitioning method to divide each type of indicator into three levels: low, medium, and high. Specifically, the corresponding indicators are arranged in ascending order of value, with the first and second tertiary loci serving as the level division boundaries. (1) If the index is not greater than the first tertiary point, it is determined to be a low level; (2) When the index is greater than the first tertiary locus but not greater than the second tertiary locus, it is determined to be of medium level; (3) When the index is greater than the second and third tertiary points, it is determined to be of a high level.

[0052] The purpose of using the equal-frequency partitioning method is to ensure that each level has a certain number of historical operation samples. This avoids the situation where, when using a fixed numerical range, the historical operation data is concentrated in a local range, resulting in insufficient samples in some state levels, which in turn affects the subsequent state transition probability statistics and the construction of the state sample library.

[0053] For the source-load correlation index, low, medium, and high levels represent different intervals of gradual change from a negative or weak correlation state to a positive correlation state. The specific physical meaning of the levels is determined by the actual values ​​corresponding to the ternary loci, rather than pre-interpreting "low level" as a negative correlation state. The sampling time... The corresponding active power load level, renewable energy output level, and source-load correlation level are combined in a fixed order to obtain the source-load operating status:

[0054] In the formula, Indicates the sampling time The corresponding source load operating status; Indicates the active load level; Indicates the level of new energy output; This indicates the source-load correlation level. The same level coding rule is used at different sampling times to ensure that the source-load operating status corresponding to different historical operating samples has a consistent meaning. For example, low, medium, and high levels can be coded as 1, 2, and 3 respectively, so the state of "low load - high renewable energy output - low source-load correlation" can be represented as... .

[0055] By combining the three levels to form the source-load operating status, historical operating conditions with different load demands, renewable energy output, and source-load relationships can be distinguished without excessively increasing the state dimensionality. The source-load operating status only describes the basic source-load operating conditions corresponding to the historical operating sample and does not directly indicate whether the historical operating sample has a safe access risk. The source-load operating status obtained in this step will be combined with the risk marker in step S302 to form a joint operating status.

[0056] S103. Perform density clustering based on the operational characteristics corresponding to historical operational samples to form operational sample clusters; Step S102 divides the historical operating samples into states according to the active load level, the output level of new energy sources and the source-load correlation. However, the state division mainly reflects the source-load conditions and cannot determine whether a certain historical operating sample belongs to a low-frequency operating state in the entire operating feature space.

[0057] Therefore, this step uses a joint feature set. The joint feature vector in the data is used as the clustering object. The operating characteristics such as distributed power output, load power, node voltage, line load rate, distribution transformer load rate and new main body operating power are used to perform density clustering on historical operating samples.

[0058] This embodiment employs the DBSCAN density clustering method. The DBSCAN method does not require pre-specifying the number of running sample clusters and can identify samples that have not formed stable local density regions, making it suitable for forming regular running sample clusters and identifying peripheral low-density running samples. Before performing clustering, continuous running variables participating in clustering are selected from the joint feature vector, and the standardized values ​​obtained during the data preprocessing stage are used to calculate the sample distance.

[0059] Let the dimension of the actual features involved in density clustering be... In this embodiment, the minimum neighborhood sample number is... Set as This configuration ensures that the core sample has a sufficient number of neighboring samples relative to the feature dimension, reducing the possibility of a small number of closely spaced samples being misidentified as a stable cluster. After determining the minimum number of neighboring samples, each joint feature vector is calculated in relation to its corresponding dimensionality. The distances between each nearest neighbor sample are calculated and arranged in ascending order to form a nearest neighbor distance curve. The distance corresponding to the inflection point where the nearest neighbor distance curve changes from a gradual change to a significant increase is selected as the neighborhood radius. The flattened portion of the nearest neighbor distance curve corresponds to a high-density region with small internal sample spacing, while the significantly increased portion corresponds to a region with decreasing sample density. Therefore, by using the inflection point to determine the neighborhood radius, it is possible to distinguish between the main operating sample cluster and the peripheral low-density samples.

[0060] For joint eigenvectors Its neighborhood sample set is represented as:

[0061] In the formula, Represents the joint feature set Another joint eigenvector in; This represents the Euclidean distance between two standardized joint feature vectors. Indicates Centered on, with a neighborhood radius of The neighborhood sample set.

[0062] When joint eigenvectors satisfy:

[0063] At that time, it was identified as the core sample.

[0064] After identifying the core samples, any core sample that has not yet been assigned to the running sample cluster is used as the starting sample, and samples located within it are... Samples within the neighborhood are grouped into the same running sample cluster; when the neighborhood samples also include other core samples, the neighborhood of the core sample is further expanded, and samples that are directly density-reachable or density-reachable are grouped into the same running sample cluster, until no new density-reachable samples exist.

[0065] Process all core samples sequentially in the manner described above to obtain the set of running sample clusters. :

[0066] In the formula, Indicates the set of sample clusters being run; Indicates the first One running sample cluster; Indicates the running sample cluster index; This indicates the total number of sample clusters being run.

[0067] Each running sample cluster is used to characterize historical running patterns that recur in the joint feature space and have stable local sample densities. This step outputs the clustering classification results of each historical running sample and inputs these results into step S104 to identify low-density running samples.

[0068] S104. Historical running samples that are not classified into any running sample cluster are identified as low-density running samples; After completing the density expansion in step S103, the joint feature set is... The historical running samples include samples that have been assigned to a running sample cluster and samples that have not been assigned to any running sample cluster.

[0069] Historical operating samples that have been assigned to any operating sample cluster have a sufficient number of neighboring samples in the joint feature space, indicating that the corresponding operating conditions recur in the historical data. Historical operating samples that have not been assigned to any operating sample cluster have fewer similar samples around them, failing to form stable operating sample clusters that meet the neighborhood density requirements. Therefore, historical operating samples that have not been assigned to any operating sample cluster are identified as low-density operating samples, resulting in a low-density operating sample set.

[0070] In the formula, This represents a low-density set of running samples.

[0071] It should be noted that unclustered samples in the DBSCAN clustering results only indicate that the sample has a low local density in the joint feature space, and cannot be directly identified as erroneous data or invalid noise. Some low-density operating samples may be formed by uncorrected anomalous measurements; others may correspond to real but historically infrequent reverse power flow, voltage overruns, equipment overload, or rapid load ramp-up conditions. Directly deleting all low-density operating samples may also delete low-frequency risk operating information with safety analysis value.

[0072] Therefore, this step only identifies low-density running samples, and does not directly delete the low-density running samples, but rather sets the low-density running sample collection together. Enter step S2 to perform data validity and risk assessment.

[0073] S2. The low-density operation samples that meet the preset data validity conditions and preset risk discrimination conditions are identified as risk operation samples; This step first uses abnormal measurement records generated during the preprocessing of historical operational data to determine the validity of low-density operational samples; then, it extracts security access risk indicators from valid low-density operational samples to identify risky operational samples with real risk operational characteristics.

[0074] In this embodiment, step S2 includes the following steps: S201. Perform data validity assessment on low-density operation samples to obtain valid low-density operation samples; During the historical operational data preprocessing stage, valid value ranges are set for each operational variable based on the rated parameters of the corresponding equipment and the effective range of the measuring device. When an operational variable exceeds its corresponding valid value range, it is marked as an abnormal measurement value. For abnormal measurement points where valid measurement values ​​exist both before and after the abnormal location, linear interpolation correction is performed using the valid measurement values ​​before and after the abnormal location, and the corresponding correction results are recorded. For data that is continuously abnormal or where there are no valid measurement values ​​before and after the abnormal location, and interpolation correction cannot be completed, the corresponding abnormal marker is retained.

[0075] For each low-density operational sample obtained in step S104, the data preprocessing record at the corresponding sampling time is read. When all operational variables in the low-density operational sample are within their corresponding valid value ranges, or when the abnormal measurement values ​​have been corrected by linear interpolation, the low-density operational sample is determined to meet the preset data validity conditions and is identified as a valid low-density operational sample. When there are still abnormal measurement values ​​in the low-density operational sample that have not been corrected, it is determined that it does not meet the preset data validity conditions and will no longer participate in subsequent risk identification, risk marking, joint operational state construction, and Markov state transition modeling.

[0076] It should be noted that, to avoid mistaking two discontinuous sampling moments before and after an invalid low-density running sample as adjacent sampling moments, only adjacent state pairs with continuous sampling times and corresponding valid historical running samples are counted when extracting state transition relationships. The sampling moment of the invalid low-density running sample forms a time breakpoint in the historical joint running state sequence, and state transition relationships are not established across this time breakpoint. Meanwhile, historical running samples that have been assigned to any running sample cluster and meet the aforementioned basic numerical availability requirements, as well as low-density running samples determined to be valid in step S201, together constitute valid historical running samples. Subsequent joint running state construction, historical joint running state sequence formation, and Markov state transition modeling all use these valid historical running samples as the processing objects.

[0077] This step directly utilizes the abnormal measurement markers and correction records generated during the data preprocessing stage to determine data validity, without setting other data validity determination methods in parallel, thereby avoiding complicating auxiliary data quality processing.

[0078] S202. Extract risk indicators from effective low-density operation samples based on security access risk-related data, and determine risk operation samples based on risk indicators and preset risk discrimination conditions. For effective low-density operating samples, read the node voltage, line load rate, distribution transformer load rate, active power at the feeder head, total active load, and total active power output of new energy sources at the corresponding sampling time.

[0079] This embodiment uses voltage over-limit indication, line overload indication, distribution transformer overload indication, reverse power flow indication, and net load ramp over-limit indication to describe the safety access risks under the new main access conditions. First, it defines the active power at the feeder head end flowing from the upstream power grid to the target distribution network as positive, and defines the reverse power flow. for:

[0080] In the formula, Indicates the sampling time The corresponding active power at the substation or feeder head end; when When the value is negative, it indicates that the target distribution network is sending active power back to the upper-level grid, and its absolute value is taken as the reverse power flow.

[0081] Net load and net load ramp are expressed as follows:

[0082]

[0083] In the formula, Indicates the sampling time Corresponding net load; Indicates the sampling time The change in net load between adjacent sampling times relative to the previous sampling time; This represents the equivalent power of the energy storage device to the net load, taking a positive value during charging and a negative value during discharging. This represents the equivalent power of an electric vehicle to its net load, taking a positive value when charging and a negative value when discharging into the grid; This represents the equivalent power of the adjustable load response relative to the net load. It is taken as a positive value when the load increases and a negative value when the load decreases. When the target distribution network is not connected to the corresponding type of new entity, the corresponding equivalent power is set to 0; when the relevant power has already been included in the total active load or the total active power output of new energy sources, it will not be added again.

[0084] Subsequently, the five risk indicators were extracted in the following order: (1) Check all node voltages. When any node voltage is lower than the preset lower limit or higher than the preset upper limit, the voltage over-limit indicator will be activated. Set to 1 otherwise set to 0, that is:

[0085] (2) Check the load rate of all lines. When the load rate of any line exceeds the preset load rate limit of the corresponding line, the line overload indicator will be adjusted. Set to 1 otherwise set to 0, that is:

[0086] (3) Check the load rate of all distribution transformers. When the load rate of any distribution transformer exceeds the preset upper limit of the load rate, the overload indicator of the distribution transformer will be activated. Set to 1 otherwise set to 0, that is:

[0087] (4) Compare the reverse power flow with a preset reverse power flow threshold. When the reverse power flow exceeds the threshold, the reverse power flow indicator value is adjusted. Set to 1 otherwise set to 0, that is:

[0088] (5) Compare the net load ramp-up amount with the preset net load ramp-up threshold. When the net load ramp-up amount exceeds the threshold, the net load ramp-up over-limit indicator is activated. Set to 1 otherwise set to 0, that is:

[0089] In the formula, This indicates an indicator function that takes the value 1 if the condition within the parentheses is true, and 0 otherwise. , , These represent the set of nodes, the set of lines, and the set of distribution transformers, respectively. Indicates the first Each node sampling time The corresponding voltage amplitude; and These represent the lower limit and upper limit of the node voltage, respectively; Indicates the first Load rate of each line; Indicates the first The maximum load rate of each line; Indicates the first The load rate of the distribution transformer; This indicates the upper limit of the load rate of the distribution transformer; Indicates the reverse power flow threshold; This indicates the net load ramp-up threshold.

[0090] Among them, the upper and lower limits of node voltage, the upper limit of line load rate, and the upper limit of distribution transformer load rate are determined according to the target distribution network operation procedures and equipment operation limits; the reverse power flow threshold is determined according to the reverse power allowed by the upstream substation or feeder; and the net load ramping threshold is determined according to the net load change capability that the target distribution network can withstand in adjacent sampling periods.

[0091] In this embodiment, five risk indicators reflect different types of security access risks. To avoid judging based on only a single risk indicator, and to retain the differences in the increase in risk severity when multiple risk conditions are triggered simultaneously, the risk indicators are weighted to obtain a risk score:

[0092] In the formula, Indicates the sampling time The corresponding risk score for effective low-density operating samples; , , , and These represent the weights of the corresponding risk indicators. This embodiment uses an equal-weighting system to ensure that the five types of risk indicators have the same initial discrimination function, i.e.:

[0093] Preset risk scoring threshold Setting it to 0.2 ensures that valid low-density operational samples triggered by at least one risk indicator can be identified as risky operational samples. When multiple risk indicators are triggered simultaneously, the risk score increases accordingly. Risky operational sample set. Represented as:

[0094] For low-density operational samples that meet the data validity criteria but have a risk score below the risk score threshold, they will be retained as low-density normal operation samples and will not be deleted as invalid data.

[0095] This step outputs a risk operation sample set. This provides a basis for step S301 to generate risk labels corresponding to valid historical running samples.

[0096] S3. Based on the source load operating status and the labeling results of the risk operating samples, determine the joint operating status, form a historical joint operating status sequence according to the sampling time, and construct a Markov state transition model based on the historical joint operating status sequence. This step combines the identification results of source load operating status and risk operating samples, so that each valid historical operating sample has both source load operating condition attributes and risk attributes. Then, a historical joint operating status sequence is formed according to the sampling time order, and the state transition patterns under different time attribute conditions are statistically analyzed.

[0097] In this embodiment, step S3 includes the following steps: S301. Generate corresponding risk labels based on the risk assessment results of each historical operation sample; The risk operation sample set only records the joint feature vector of samples identified as risky operations, but Markov state transition modeling requires determining a uniform risk attribute for each valid historical operation sample participating in the modeling. Therefore, for each valid historical operation sample, the joint feature vector... Determine whether it belongs to the risk operation sample set and generate a risk label:

[0098] In the formula, Indicates the sampling time The corresponding risk marker. Among them, when... When it belongs to the risk operation sample set, the risk label is 1; when If a sample does not belong to the risk operation sample set, the risk label is set to 0.

[0099] Therefore, regular operation samples, low-density normal operation samples, and risky operation samples all have a unified risk attribute expression. Among them, the risk label of low-density normal operation samples is 0, and they still participate in the subsequent joint operation status construction and status sample library collection.

[0100] S302. Combine the source load operation status and risk markers corresponding to each historical operation sample to obtain the joint operation status of each historical operation sample; The source-load operating status describes the load level, renewable energy output level, and source-load correlation corresponding to historical operating samples, but it cannot distinguish between normal samples and risk samples under the same source-load conditions. For example, historical operating samples with the same "low load - high renewable energy output - low source-load correlation" state may exhibit normal operating status and reverse power flow or voltage over-limit risk status respectively due to different local network power distributions. Therefore, the sampling time... The corresponding source load operating status and risk markers are combined:

[0101] In the formula, Indicates the sampling time The corresponding joint operation status.

[0102] Through the above combination, historical operating samples with the same source-load operating status but different risk labels are divided into different joint operating states, enabling risk attributes to participate in subsequent state transition modeling. Deduplication of the joint operating states corresponding to all valid historical operating samples yields the set of joint operating states:

[0103] In the formula, Represents the set of joint operating states; Indicates the first Class union runtime state; Indicates the category index of joint operation status; This represents the total number of joint operational status categories that actually occurred in historical data.

[0104] By including only the joint operating states that actually occur in historical data into the state set, we can avoid creating empty states without historical samples and reduce the dimension of the subsequent state transition matrix.

[0105] S303. Arrange the joint operation status according to the sampling time to form a historical joint operation status sequence, and divide the historical operation samples with the same joint operation status into the corresponding status sample library; To statistically analyze the evolution of the distribution network's operating status over time, this embodiment arranges the joint operating states corresponding to each valid historical operating sample in chronological order of sampling time, forming a historical joint operating state sequence:

[0106] In the formula, Arranged from first to last according to sampling time, these represent the valid historical sampling moments that participated in the construction of the joint operational status. This represents the position index in the valid historical sampling time sequence, and ; This represents the number of valid historical running samples. The use of a valid historical sampling time sequence is to preserve time breaks caused by the exclusion of invalid low-density running samples. Subsequent iterations will only be performed when... Only then will and This is used as a group of adjacent joint operating states for state transition statistics. While forming the historical joint operating state sequence, valid historical operating samples are aggregated according to the joint operating state category. For the... The class union runtime state will satisfy The joint feature vectors are divided into the corresponding state sample database:

[0107] In the formula, Indicates the first A database of state samples corresponding to the class's joint running state.

[0108] It should be noted that the reason for establishing the state sample library is that the Markov state transition model generates a discrete joint operating state sequence. This state sequence can only represent the state category at each target sampling time of the target period, and cannot directly provide the specific values ​​of continuous operating variables such as distributed power output, load power, new main body operating power, and node voltage. The state sample library saves valid historical operating samples that have actually occurred under each joint operating state, so that after generating the joint operating state sequence for the target period, it can call the actual operating samples with corresponding source-load operating condition attributes and risk attributes according to the state category. Each state sample library retains the original operating variables, source-load correlation matrix, risk indicator, and risk score of the historical operating samples; at the same time, it retains the mean and standard deviation of each operating variable determined at the beginning of the embodiment for standardization processing during subsequent scenario distance calculation.

[0109] S304. Extract the state transition relationship between adjacent joint operation states from the historical joint operation state sequence; The historical joint operation state sequence records the state evolution order of the target distribution network over a continuous historical period. This embodiment constructs a first-order Markov state transition model, taking the joint operation state at the current sampling moment as the current state and the joint operation state at the next consecutive sampling moment as the successor state. State pairs in the historical joint operation state sequence are read sequentially. For any two consecutive valid historical sampling moments... and When the following conditions are met:

[0110] At that time, will:

[0111] It was determined to be a group of adjacent joint operating states.

[0112] when ,and At that time, record one instance of joint operation status. To joint operation status The state transition.

[0113] in, Indicates the current joint operation status category index; Indicates the next joint running state category index, and:

[0114] It should be noted that, and Each value can be taken independently. For example, the current joint operating state can be determined by... Transfer to It is not required that the category number of the next joint operation state is greater than the category number of the current joint operation state.

[0115] when If a state transition relationship is not established between two valid historical running samples, it indicates that there are invalid samples or missing sampling times between them. After traversing all consecutive valid sampling times, the historical transition relationships and corresponding transition counts between different joint running states are obtained.

[0116] Since risk markers are already part of the joint operating state, invalid samples or missing sampling times exist among the state transition samples, and no state transition relationship is established between them. The traversal relationship can simultaneously reflect the changes between different source load operating states, the transition from normal joint operating state to risky joint operating state, the recovery from risky joint operating state to normal joint operating state, and the continuous transition between different risky joint operating states.

[0117] S305. Classify the state transition relationships according to the time attributes corresponding to the historical operation samples, and calculate the state transition probabilities under different time attributes. The operating status of a power distribution network exhibits a clear time dependence. The probability of the same combined operating state occurring subsequently may differ depending on the time of day, day type, and season. For example, a high-energy-output state is more likely to maintain high output during the day, but the probability of shifting to a medium-to-low energy-output state increases significantly near sunset; load variation patterns differ between weekdays and holidays; and the energy output and load characteristics also vary across different seasons.

[0118] Therefore, based on the sampling time of each valid historical running sample, hourly attributes, weekday attributes, holiday attributes, and seasonal attributes are extracted, and time attribute categories are formed by combining the time attributes.

[0119] sampling time The corresponding time attribute category is denoted as For adjacent joint operation states The sampling time corresponding to the subsequent joint operation status Time attribute category The time attribute category is used as the classification category for this state transition, and the state transition samples obtained in step S304 are classified accordingly. When the time attribute category is... Under the condition of joint operation status Transfer to joint operation status The state transition probability is expressed as:

[0120] In the formula, each summation is performed for the time specified at the sampling time. and This is performed using consecutive historical transfer samples that are all valid in their historical operation. The time attribute category indicating the subsequent sampling time is: At that time, from the joint operation status Transfer to joint operation status The local state transition probability.

[0121] Wherein, the numerator of the formula represents the time attribute category. And actually happened The number of historical transfer samples; the denominator of the formula represents the corresponding time attribute category, with... This represents the total number of consecutive valid historical transfer samples in the current joint operation state. A denominator of zero indicates that there are no samples in the corresponding time attribute category that are in the joint operation state. As a historical transition sample of the current state, the local state transition probability is not calculated at this time, and the global state transition probability is used in step S306.

[0122] S306. Based on the number of historical transition samples under different time attributes, the state transition probability is corrected using the global state transition probability, and a state transition matrix is ​​constructed based on the corrected state transition probability to form the Markov state transition model. After classifying the state transition samples according to their time attributes, some low-frequency joint operation states, especially those with a risk label of 1 (risk label in step S301), were identified. (Option 1) Under certain time attribute categories, there may only be a small number of historical transition samples. Directly using a small number of historical samples to calculate the local state transition probability could easily assign an excessively high probability to a single, accidental transition, or assign a value of zero to transitions that actually occurred under other time conditions in the past, making it difficult for low-frequency risk states to enter the joint operating state sequence of the target period. Therefore, while calculating the local state transition probability, without distinguishing between time attribute categories, the global state transition probability is calculated using all continuous and valid historical transition samples:

[0123] In the formula, This represents the combined operational status obtained from all consecutive valid historical transfer samples. Towards joint operation status The global state transition probability of the transition.

[0124] This embodiment addresses each current joint operation state. and each time attribute category Statistics based on the current joint operation status The departure time and subsequent sampling time have time attribute categories Number of historical transfer samples:

[0125] In the formula, Indicates that the time attribute category is At that time, in joint operation status This represents the number of historical transition samples for the current state. This number of historical transition samples is used to determine the state transition matrix. Does the local state transition probability corresponding to a row have sufficient historical sample support?

[0126] Pre-set historical transfer sample number threshold and according to Determine the smoothing coefficient:

[0127] By smoothing the local state transition probabilities using the global state transition probabilities, we obtain:

[0128] Among them, when season Instead of calculating the corresponding local state transition probabilities separately, the global state transition probabilities are used directly; when When, the local state transition probability and the global state transition probability are weighted according to the smoothing coefficient; when season The local state transition probability is used.

[0129] Therefore, when At that time, with the change from the current joint operation status The departure time and subsequent sampling time have time attribute categories As the number of historical transition samples increases, the weight of the corrected state transition probability on the local state transition probability increases accordingly; conversely, as the number of historical transition samples decreases, the weight of the corrected state transition probability on the global state transition probability increases accordingly. When When the state transition probability is corrected, the local state transition probability under the corresponding time attribute category is adopted.

[0130] When the current joint operation status If a state does not appear in any of the continuous valid historical transition samples, making it impossible to calculate the corresponding global state transition probability, then the current joint operating state will be used. The probability of transitioning to itself is set to 1, and the probabilities of all other transitions are set to 0, to ensure that the sum of the probabilities in the corresponding rows of the state transition matrix is ​​1. The corrected state transition probabilities are then arranged according to the order of states in the joint running state set to construct time attribute categories. The corresponding state transition matrix :

[0131] In the formula, the state transition matrix is ​​the first... Line 1 The elements in the column represent the time attribute category. Under the following conditions, the joint operation status Transfer to joint operation status The probability of.

[0132] Therefore, the state transition matrices corresponding to different time attribute categories collectively constitute the Markov state transition model. Through the aforementioned smoothing correction, the situation where low-frequency joint operating states are assigned zero transition probability due to insufficient local samples can be reduced, allowing historically existing risky joint operating states to participate in the generation of the target time period's state sequence.

[0133] S4. Generate a joint operation state sequence for the target time period based on the Markov state transition model, and generate a low-probability risk scenario for the distribution network containing risk operation characteristics based on the joint operation state sequence and the corresponding operation sample. This step first calls the corresponding state transition matrix based on the time attributes of each target sampling time within the target time period to generate the joint operating state sequence of the target time period; then, basic samples are selected from the corresponding state sample library, and multivariate candidate scenarios are generated through correlation-preserving perturbation reconstruction and operating constraint verification; finally, the multivariate candidate scenarios are clustered and screened to determine representative scenarios that contain risky operating characteristics and have a low probability of occurrence.

[0134] In this embodiment, step S4 includes the following steps: S401. Obtain the time attributes corresponding to each sampling time within the target time period and match the corresponding state transition matrix; The target time period is determined based on the required time range for secure access analysis, and the sampling period of the target time period is kept consistent with the sampling period of historical operational data. In this embodiment, a continuous 24-hour period is used as the target time period, with a sampling period of 15 minutes, thus the target time period includes 96 target sampling moments. The length of the target time period is consistent with the length of the correlation calculation time period used in step S101, enabling the subsequent calculation of the corresponding source-load correlation matrix based on the source-load sequence of the complete target time period.

[0135] In the process of generating the scene for the target time period, the following methods are adopted: Represents the index of the target sampling time within the target time period, and:

[0136] In the formula, This represents the total number of target sampling times within the target time period, in this embodiment. .

[0137] The target sampling time in this step Compared with the aforementioned historical sampling time and The timeframes they occupy are different. Among them, and Used for historical operational data processing and model building. This is used for generating state sequences and scenarios for the target time period. Historical operating data and the target time period use the same sampling period to ensure that the state transition matrix, state sample library, and target scenario correspond at the same time resolution. For each target sampling moment within the target time period, the corresponding hourly, weekday, holiday, and seasonal attributes are extracted according to the same classification rules in step S305 and combined to obtain the target time attribute category. Subsequently, based on the target time attribute category corresponding to each target sampling moment, the corresponding state transition matrix is ​​called from the Markov state transition model formed in step S306. The called state transition matrix is ​​used to generate the joint operating state corresponding to that target sampling moment. When a target time attribute category does not form an independent state transition matrix, a global state transition matrix that does not distinguish between time attributes is called to ensure that the state sequence for the target time period can be continuously generated.

[0138] S402. Take the initial joint operation state of the target time period as the current joint operation state, perform state transition sampling through the matched state transition matrix, and generate the joint operation state sequence of the target time period hour by hour. The joint operation state corresponding to the actual sampling time before the start of the target time period is taken as the 0th joint operation state, denoted as... For the first Each target sampling time, Call the time attribute category corresponding to the sampling time of the target. The corresponding state transition matrix When the current joint operation status At that time, with As the row index of the state transition matrix, the data is read from the joint running state. The state transition probabilities to each candidate next joint operating state are determined, and sampling is performed accordingly. .

[0139] According to the state arrangement order in the joint operation state set, for the first... The state transition probabilities of each row are accumulated sequentially, forming a cumulative probability interval covering 0 to 1. Then, a random number between 0 and 1 is generated. Based on the cumulative probability interval into which this random number falls, the joint operating state for the next target sampling time is determined. This generated next joint operating state is used as the new current joint operating state. Then, based on the time attribute of the next target sampling time, the corresponding state transition matrix is ​​called, and the state transition sampling process is repeated until the joint operating states for all target sampling times within the target time period are obtained. Then the th... The joint operational status sequence of the target time period is represented as follows:

[0140] In the formula, Indicates the first Target sampling time in a state sequence The corresponding joint operation status; This indicates the sequence number of the joint operation status during the target time period.

[0141] Because state transition sampling is random, repeating the above process can yield multiple joint operating state sequences that conform to the same time attribute state transition law but have different state evolution paths, so as to cover multiple state evolution processes that may occur in the target time period.

[0142] S403. For the joint operation state sequence of the target time period, resample from the state sample library of the corresponding joint operation state to form a basic scenario; The target time period joint operation state sequence obtained in step S402 consists of discrete joint operation states, which can indicate the source and load operating conditions and risk attributes corresponding to each target sampling time within the target time period, but do not yet provide specific values ​​for distributed power output, load power, new main body operating power, and distribution network operation state quantities. Therefore, for each joint operation state in the target time period joint operation state sequence, actual historical operation samples are called from the corresponding state sample library established in step S303 to transform the discrete joint operation states into basic samples with specific operating variables.

[0143] For the A joint operating state sequence, when the target sampling time The corresponding joint operation status is When the basic sample selection relationship is expressed as:

[0144] In the formula, Indicates the first The joint operation status sequence of the target time period at the target sampling time The corresponding joint operation status; Indicates joint operation status The corresponding state sample library; Indicates from the state sample library The basic joint feature vector extracted from it.

[0145] Due to the state sample library All historical running samples in the database share the same joint running state. Therefore, the basic joint feature vector extracted from this state sample library can inherit the joint running state corresponding to the target sampling time at the level of specific running variables. This embodiment adopts a resampling method with replacement, that is, after completing one sample extraction, the extracted historical running samples are still retained in the corresponding state sample library for continued use in subsequent target sampling times or other scenario generation processes. The reason for using resampling with replacement is that the joint running state of the corresponding risky running sample usually appears less frequently in the historical running data, and the number of historical running samples in its state sample library may be less than that of regular joint running states. If sampling without replacement is used, it is easy to lead to insufficient number of callable samples of low-frequency joint running states when generating multiple basic scenarios, thereby limiting the number of scenarios containing risky running features generated.

[0146] The basic joint feature vectors obtained by resampling all target sampling times within the target time period in sequence and arranging them according to the chronological order of the target sampling times are formed into the first... Basic scenarios:

[0147] In the formula, Indicates the first A basic scene sequence.

[0148] Each basic joint feature vector in the basic scenario originates from actual historical running samples and matches the joint running state at the corresponding target sampling time. Therefore, the basic scenario has a foundation of actual historical running data, but it is essentially a recombination of historical running samples and needs to be further perturbed and reconstructed in step S404 to form a new scenario different from the existing historical samples.

[0149] S404. Based on the correlation between the running variables in the corresponding state sample library, the basic scenario is reconstructed by correlation-preserving perturbation. If the basic scenario formed in step S403 is directly used as the final generated scenario, the resulting scenario is merely a rearrangement of historical operating samples, making it difficult to represent other reasonable combinations of operating variables that may occur under the same joint operating state. Therefore, this step, while preserving the multivariate joint change relationships in the corresponding state sample library, perturbs and reconstructs the basic scenario, so that the reconstructed scenario is neither different from the original historical samples, nor does it destroy the correlation between distributed power output, load power, and other operating variables.

[0150] For the target sampling time in the basic scenario The corresponding basic joint feature vector First, determine the state sample library based on its corresponding joint operation state. Subsequently, based on the state sample library... For each historical operating sample, the continuous operating variables are used to determine the multivariate central value corresponding to the state sample library, and the joint deviation of each historical operating sample relative to the multivariate central value is calculated. Each joint deviation includes the synchronous changes of distributed power output, load power, new main unit operating power, and other continuous operating variables at the same historical sampling time. A joint deviation is randomly selected from the joint deviations corresponding to the state sample library, and the amplitude of the joint deviation is limited according to the historical fluctuation range of each operating variable in the state sample library to obtain the target sampling time. Corresponding perturbation vector Adding the perturbation vector to the basic joint feature vector yields the reconstructed candidate joint feature vector:

[0151] In the formula, Indicates the first The reconstructed scenario at the target sampling time The corresponding candidate joint feature vector; Indicates the target sampling time The corresponding perturbation vector.

[0152] The changes in each operating variable in the disturbance vector originate from the joint deviation of the same historical operating sample, rather than generating random disturbances independently for distributed power output, load power, and other operating variables. This method preserves the synchronous change direction of each operating variable in the corresponding state sample library, reducing the possibility of distortion in source-load correlation caused by independent disturbances. After disturbance processing, it is checked whether each continuous operating variable in the candidate joint feature vector falls within the historical fluctuation range of the corresponding state sample library. If any operating variable exceeds the corresponding historical fluctuation range, the current disturbance vector is discarded, and new variables are extracted from the joint deviation of the corresponding state sample library.

[0153] Candidate joint feature vectors that have passed the historical fluctuation range check are arranged in chronological order of the target sampling time to form a reconstructed scene sequence. :

[0154] Although a single disturbance vector originates from the joint deviation of historical operating samples, perturbing multiple basic joint feature vectors within the target time period may still cause the source-load correlation of the entire reconstructed scenario to deviate from the basic scenario. Therefore, based on the distributed power output sequence and load power sequence within the target time period in both the basic and reconstructed scenarios, and according to the same access point arrangement order and load area arrangement order described in steps S1011 and S1012, the scenario-level source-load correlation matrix is ​​calculated.

[0155] The condition for preserving source-load correlation between the base scenario and the reconstructed scenario is expressed as follows:

[0156] In the formula, Represents a refactoring scenario The corresponding source-load correlation matrix; Representing the basic scenario The corresponding source-load correlation matrix; Denotes the Frobenius norm; This represents the source-load correlation deviation threshold.

[0157] Among them, the difference between the source-load correlation matrices is used for verification, which can simultaneously compare the local correlation between different distributed power access points and different load areas, avoiding the situation where the average source-load correlation index of two scenarios is similar, but some local source-load correlation has changed significantly.

[0158] The source-load correlation deviation threshold is determined based on the normal fluctuation range of the source-load correlation relationship in the historical operation data of the target distribution network. Specifically, a continuous historical operation sample sequence with the same length as the target time period is selected from the historical operation data, the distribution of the source-load correlation matrix differences between each continuous historical operation sample sequence is statistically analyzed, and the source-load correlation deviation threshold is determined according to the allowed historical normal fluctuation range.

[0159] When a reconstructed scenario meets the source-load correlation preservation condition, it is input into step S405; when a reconstructed scenario does not meet the condition, the joint deviation is re-extracted and perturbation reconstruction is performed. Reconstructed scenarios that still do not meet the source-load correlation preservation condition after reaching the preset maximum number of reconstructions are no longer retained.

[0160] S405. Perform runtime constraint verification on the reconstructed scene to obtain multivariate candidate scenes; The source-load correlation preservation verification in step S404 only indicates that the output of distributed generation and the load power in the reconfigured scenario retain a similar correlation to that in the basic scenario. It cannot directly indicate that the reconfigured multivariate power combination meets the actual operational requirements of the target distribution network and various new types of entities. Therefore, the output of distributed generation, load power, energy storage charging and discharging power, electric vehicle charging and discharging power, and adjustable load response power at each target sampling time in the reconfigured scenario are converted into node power injection according to the actual access location of each device in the target distribution network, and the power flow calculation of the target distribution network is performed sequentially.

[0161] Based on the power flow calculation results, perform constraint verification in the following order: First, check whether the voltage of each node is within the physically feasible voltage boundary of the target distribution network and related equipment, and check whether the load rate of each line and the load rate of the distribution transformer exceed the physical limit load rate of the corresponding equipment. The node voltage constraint, line load rate constraint, and distribution transformer load rate constraint are respectively expressed as: Node voltage:

[0162] Line load rate:

[0163] Distribution transformer load rate:

[0164] In the formula, Indicates the first In the first reconstruction scenario, the... Each node at the target sampling time The voltage amplitude; and These represent the lower limit and upper limit of the node voltage allowed by the target distribution network, respectively; Indicates the first In the first reconstruction scenario, the... The line at the target sampling time Load rate; Indicates the first The maximum allowable load rate for each line; Indicates the first In the first reconstruction scenario, the... Taiwan distribution transformer at the target sampling time Load rate; This indicates the maximum allowable load rate of the distribution transformer; , and These represent the node index, line index, and distribution transformer index, respectively.

[0165] In step S202, the node voltage upper and lower limits and the equipment load rate upper limit are used as risk assessment thresholds. In step S405, the operable boundary is the boundary corresponding to the physical capacity of the equipment and the operable range of the distribution network. The risk assessment threshold is located within the corresponding operable boundary, that is, it satisfies:

[0166]

[0167]

[0168] It should be noted that the node voltage upper and lower limits and equipment load rate upper limit in step S202 are risk assessment thresholds used to identify high-risk operating conditions that require focused analysis; the node voltage boundary and equipment load rate boundary in this step are physical feasibility boundaries used to exclude scenarios that exceed the physical capacity of the equipment or cannot form an effective power flow operation section. The risk assessment thresholds are located within the corresponding physical feasibility boundaries. Therefore, scenarios that exceed the risk assessment thresholds but do not exceed the physical feasibility boundaries are still retained.

[0169] When the voltage of each node in the reconfiguration scenario meets the node voltage constraint, the load rate of each line meets the line load rate constraint, and the load rate of each distribution transformer meets the distribution transformer load rate constraint, the output of distributed power sources and the charge status of energy storage continue to be verified; if any constraint is not met, it is determined that the current reconfiguration scenario exceeds the operable range of the target distribution network.

[0170] Secondly, check whether the active power output of each distributed power source is within the allowable output range of the corresponding device. The output constraint of the distributed power source is expressed as follows:

[0171] In the formula, Indicates the first In the first reconstruction scenario, the... Distributed power access points at the target sampling time Those who have made meritorious contributions; and They represent the first The minimum and maximum active power output allowed for each distributed power source access point; This represents the index of the distributed power supply access point.

[0172] The lower limit of active power output for distributed photovoltaic (PV) and distributed wind power is typically determined based on their minimum allowable output, while the upper limit is determined based on the rated active power capacity of the equipment and the available output at the target sampling time. This constraint prevents the output of distributed power sources after disturbance reconfiguration from exceeding the actual output capacity of the equipment.

[0173] For energy storage devices and electric vehicles with charging and discharging capabilities, check whether their charging and discharging power are within the allowable power range of the corresponding devices under the current connection state. For electric vehicles not connected to the target distribution network or in a non-charging / discharging state, set their charging and discharging power to 0. For adjustable loads, check whether their response power, continuous response duration, and cumulative response amount meet the operating constraints of the corresponding load devices. If the operating constraints of any new entity are not met, the current reconfiguration scenario is deemed not feasible.

[0174] Subsequently, based on the state of charge (SOC) at the previous target sampling time, the current charging / discharging power, the sampling period, and the charging / discharging efficiency, the SOC of the energy storage device or the electric vehicle with a SOC model is updated hourly, and it is determined whether the updated SOC is within the allowable SOC range. The SOC constraint is expressed as follows:

[0175] In the formula, Indicates the first In the first reconstruction scenario, the... An energy storage device at the target sampling time The state of charge; and They represent the first The permitted state of charge (SOC) limit and SOC upper limit for an energy storage device; the energy storage device includes energy storage units and electric vehicles equipped with a SOC model.

[0176] The energy storage state of charge (SOC) is not assigned independently for each target sampling time, but is recursively obtained according to the chronological order of the target sampling times. The actual SOC of the energy storage before the start of the target time period is used as the initial SOC. The charging and discharging power at the current target sampling time is used to update the SOC at the next target sampling time, thereby ensuring that the energy storage charging and discharging power sequence and the SOC sequence are continuous in time.

[0177] Finally, check whether the power flow calculations corresponding to each target sampling time have converged, and determine whether the power balance deviation of the distribution network is within the allowable range. The power balance constraint is expressed as:

[0178] In the formula, Indicates the first The reconstructed scenario at the target sampling time Power balance deviation; This indicates the preset power balance deviation threshold.

[0179] The power balance deviation is determined based on the difference between the total node power injection, line transmission power, and network loss at the target sampling time. When the power flow calculation fails to converge, or the absolute value of the power balance deviation exceeds the power balance deviation threshold, it indicates that the current reconfiguration scenario cannot form an effective distribution network operation section, and the reconfiguration scenario is discarded.

[0180] In summary, the refactoring scenario needs to simultaneously satisfy:

[0181] The aforementioned constraints should be verified separately for each target sampling time within the target time period. The reconstructed scenario is deemed to satisfy the operational constraints only if it satisfies the constraints at all target sampling times and the corresponding power flow calculations converge.

[0182] The risk discrimination threshold in step S202 is used to determine whether the corresponding operating condition has a security access risk that needs to be analyzed in detail; the operating constraint boundary in this step is used to determine whether the reconfiguration scenario is still within the physical capacity of the equipment and the operability of the distribution network. The two have different technical functions. Therefore, when the reconfiguration scenario exceeds the risk discrimination threshold corresponding to node voltage, equipment load rate, reverse power flow, or net load ramp-up, but has not yet exceeded the operability boundary of the equipment, the scenario and its risk operating characteristics are retained; when the reconfiguration scenario shows power flow non-convergence, equipment operating variables exceeding physical capacity, or inability to meet power balance relationships, the reconfiguration scenario is determined to be inoperable.

[0183] After completing the operational constraint verification, the source and load operational states corresponding to each target sampling time in the reconstructed scenario are redefined according to the source and load operational state division rules in step S102. The risk score corresponding to each target sampling time is calculated according to the risk indicator extraction rules and risk score calculation rules in step S202. Based on whether the risk score reaches a preset risk score threshold, a risk marker for the corresponding target sampling time in the reconstructed scenario is generated. Since low-density attributes are the distribution attributes of historical operational samples in the joint feature space, the risk marker for the reconstructed scenario is no longer determined based on whether it belongs to the historical low-density operational sample set, but rather on whether it possesses the corresponding risk operational characteristics. The redefined source and load operational states and risk markers are combined to obtain the reconstructed joint operational state.

[0184] Specifically, when calculating the source-load correlation corresponding to the beginning of the target time period, the continuous historical operating data before the start of the target time period is connected with the already generated target time period data to form an output sequence and a load sequence that meet the preset correlation calculation time period length.

[0185] The reconstructed joint operating state is compared hourly with the target time period joint operating state sequence generated in step S402. When the joint operating state of the reconstructed scenario at each target sampling time is consistent with the target time period joint operating state sequence, it indicates that the reconstructed scenario retains the source-load operating conditions and risk attributes defined by the target state sequence. When the reconstructed joint operating state is inconsistent with the target time period joint operating state sequence, the process returns to step S404 for perturbation reconstruction; scenarios that still cannot maintain state consistency after reaching the preset maximum number of reconstructions are discarded. Scenarios that simultaneously satisfy the source-load correlation preservation condition, operating constraint condition, and joint operating state consistency condition are identified as multivariate candidate scenarios, forming a multivariate candidate scenario set.

[0186] In the formula, Represents a set of multivariate candidate scenarios; Indicates the first A multivariate candidate scenario.

[0187] S406. Based on the operational characteristics, source-load correlation characteristics, and risk characteristics of multivariate candidate scenarios, construct a comprehensive scenario distance between multivariate candidate scenarios; The multivariate candidate scenarios obtained in step S405 simultaneously include distributed power generation output, load power, new main body operating power, distribution network operating status variables, source-load correlation, and security access risk characteristics. If scenario clustering is performed solely based on the numerical differences of each operating variable in the candidate scenarios, scenarios with similar operating power levels but significantly different source-load correlation or risk levels may be grouped into the same candidate scenario cluster. If clustering is performed solely based on risk level, the differences in source-load operating conditions and time-series changes among the scenarios may be ignored. Therefore, this embodiment calculates the operating characteristic distance, source-load correlation distance, risk characteristic distance, and time-series morphology distance between multivariate candidate scenarios, and normalizes and weights each distance term to obtain the comprehensive scenario distance.

[0188] For any two multivariate candidate scenarios and ,have:

[0189] In the formula, and These represent the indices of the two multivariate candidate scenarios.

[0190] S4061. Calculate the running feature distance, as follows: The operational characteristic distance characterizes the overall differences between two multivariate candidate scenarios in terms of distributed generation output, load power, new main body operating power, and distribution network operating state variables within the target time period. Since the operational variables involved in the distance calculation have different dimensions and numerical ranges, each continuous operational variable is standardized according to the mean and standard deviation determined during the historical operational data preprocessing stage before calculating the operational characteristic distance. For multivariate candidate scenarios... At the target sampling time The corresponding number Class runtime variables, their standardized values ​​are represented as:

[0191] In the formula, Representing multivariate candidate scenarios At the target sampling time The corresponding number The original value of the class's runtime variable; This represents the corresponding standardized value; and They represent the first The mean and standard deviation of the class's runtime variables over a preset historical period; This indicates the category index of the runtime variable involved in the distance calculation. The mean and standard deviation are consistent with those used in the historical runtime data preprocessing stage to ensure that similar runtime variables in historical runtime samples and generated scenarios have the same standardized scale. When the standard deviation corresponding to a certain category of runtime variable is zero, the standardized value of that runtime variable is set to zero.

[0192] The standardized operational variables involved in the calculation of operational feature distance at the same target sampling time are arranged in a fixed order to form a standardized operational feature vector, denoted as [vector]. The standardized running feature vector is only used as an intermediate calculation result in this step and is not saved as a new candidate scene or sample set.

[0193] Therefore, the running feature distance between two multivariate candidate scenarios is expressed as:

[0194] In the formula, Representing multivariate candidate scenarios and Distance between operational features; and These represent the two multivariate candidate scenarios at the target sampling time. The corresponding standardized operational feature vector. Averaging the differences in operational features across all target sampling times provides a comprehensive reflection of the numerical differences in operational variables between two multivariate candidate scenarios throughout the entire target time period, avoiding comparisons based solely on a single target sampling time while ignoring the overall operational level of the scenario.

[0195] S4062. Calculate the source-load correlation distance, as follows: While operational characteristic distances can reflect the overall differences in operational variable values, they cannot fully characterize the differences in source-load correlation between each distributed power source access point and each load region. Therefore, based on the distributed power source output sequence and load power sequence of each multivariate candidate scenario within the target time period, and following the same calculation method, order of distributed power source access points, and order of load regions as in steps S1011 and S1012, the corresponding source-load correlation matrices are calculated respectively. Multivariate Candidate Scenarios and The source-load correlation distance between them is expressed as:

[0196] In the formula, Representing multivariate candidate scenarios and The source-load correlation distance between them; and These represent the source-load correlation matrices calculated based on two multivariate candidate scenarios. The source-load correlation matrices corresponding to different multivariate candidate scenarios use the same matrix row and column order; therefore, matrix elements at the same position in both matrices correspond to the same distributed power supply access points and load areas. By calculating the Frobenius norm of the difference between the two source-load correlation matrices, the differences between all corresponding matrix elements can be comprehensively reflected. Compared to comparing only the source-load correlation index, this processing can preserve the differences in local source-load correlation relationships, avoiding situations where the average source-load correlation levels of the two scenarios are similar, but the correlation relationships between some distributed power supply access points and load areas are significantly different.

[0197] S4063. Calculate the risk feature distance, as follows: According to the risk indicator extraction rules and risk score calculation rules in step S202, calculate the risk score corresponding to each multivariate candidate scenario at each target sampling time. Multivariate candidate scenarios The average scenario risk score during the target period is expressed as follows:

[0198] In the formula, Representing multivariate candidate scenarios Average scenario risk score during the target period; Representing multivariate candidate scenarios At the target sampling time The corresponding risk score. The average scenario risk score is used to characterize the overall risk level of the candidate scenario throughout the entire target time period. When the duration of the risk operation state in the candidate scenario is long, or when multiple target sampling times simultaneously trigger a large number of risk indicators, its average scenario risk score increases accordingly. The risk feature distance between two multivariate candidate scenarios is expressed as:

[0199] In the formula, Representing multivariate candidate scenarios and The risk feature distance between the two scenarios is calculated by taking the absolute value of the difference between the average scenario risk scores. This ensures that the risk feature distance is non-negativity and symmetry, meaning that changing the order of the two multivariate candidate scenarios will not change the risk feature distance between them.

[0200] It should be noted that the average scenario risk score in this step is used to compare the overall risk level of different candidate scenarios; the subsequent step S408 determines whether the representative scenario contains risk operation characteristics based on whether there is a target sampling time that meets the risk discrimination conditions in the representative scenario. The two have different technical uses.

[0201] S4064. Calculate the temporal morphological distance, as follows: The operational characteristic distance primarily reflects the difference in operational variable values ​​between two multivariate candidate scenarios at corresponding target sampling times. However, even if two scenarios have similar overall operational levels, their power ramp-up direction, magnitude of change, and temporal fluctuation processes may still differ significantly. Therefore, the temporal morphological distance is calculated based on the standardized operational characteristic changes between adjacent target sampling times. (Multivariate candidate scenarios) At the target sampling time The corresponding standardized operating characteristic changes are expressed as follows:

[0202] In the formula, Representing multivariate candidate scenarios From the target sampling time Up to the target sampling time The standardized operational characteristic changes. Therefore, the temporal morphological distance between two multivariate candidate scenarios is expressed as:

[0203] In the formula, Representing multivariate candidate scenarios and Temporal morphological distance between them; and These represent the standardized changes in operational characteristics of two multivariate candidate scenarios between adjacent target sampling times. The smaller the temporal morphological distance, the closer the direction and magnitude of the changes in each operational variable over time in the two multivariate candidate scenarios; the larger the temporal morphological distance, the more significant the differences between the two in terms of power ramp-up, fluctuation amplitude, or temporal change rhythm.

[0204] S4065. Normalize each distance term as follows: The runtime feature distance, source-load correlation distance, risk feature distance, and time series morphology distance use different calculation objects, and their numerical ranges may differ significantly. If the four types of distances are directly weighted and combined, the distance term with the larger numerical range may have an unreasonable dominant effect on the overall scenario distance. Therefore, min-max normalization is performed on each distance term based on the maximum and minimum values ​​of the corresponding distance terms between all multivariate candidate scenarios. For any distance type... The normalized distance is expressed as:

[0205] In the formula, Indicates the distance type; and These represent the maximum and minimum values ​​of the corresponding distance type across all multivariate candidate scene pairs; This represents the normalized distance.

[0206] When a certain distance type satisfies:

[0207] When the distance type is zero, it indicates that the distance type does not have a distinguishing effect among the current multivariate candidate scenarios. The normalization result corresponding to the distance type is uniformly set to 0 to avoid the denominator being zero in the normalization calculation.

[0208] S4066. Construct a comprehensive scene distance, as detailed below: The normalized four types of distances are weighted and combined to obtain the comprehensive scene distance between multivariate candidate scenes:

[0209] In the formula, Representing multivariate candidate scenarios and The overall scene distance between them; , , and These represent the weights of the operational characteristic distance, source-load correlation distance, risk characteristic distance, and time series morphology distance, respectively. Each distance weight is non-negative and satisfies the following conditions:

[0210] This embodiment uses equal weighting for the four types of distances, namely:

[0211] By adopting an equal weighting setting, the differences in the numerical values ​​of operating variables, the differences in the correlation between source and load, the differences in the overall risk level, and the differences in time series changes are all included in the comprehensive scene distance calculation, thus avoiding the situation where a single type of distance dominates the scene clustering result when there is a lack of clear scene feature preferences.

[0212] Based on the comprehensive scene distances between all pairs of multivariate candidate scenes, a comprehensive scene distance matrix is ​​constructed, and the comprehensive scene distance matrix is ​​used as the distance input for K-medoids clustering in step S407.

[0213] S407. Cluster the multivariate candidate scenes based on the comprehensive scene distance to obtain representative scenes and their corresponding probabilities. The number of multivariate candidate scenarios obtained in step S405 is usually greater than the number of scenarios that can be directly handled in subsequent security access assessments, distribution network planning, and operation verifications. Therefore, it is necessary to reduce the number of multivariate candidate scenarios while preserving differences in operating conditions, source-load relationships, risk levels, and time series changes.

[0214] To avoid categorizing multivariate candidate scenarios with and without risky operational characteristics into the same scenario cluster, thus potentially overlooking risky scenarios due to the selection of representative scenarios, the multivariate candidate scenarios are first divided into risky candidate scenario subsets and non-risky candidate scenario subsets according to the risk indicator extraction rules and risk score calculation rules in step S202. Then, K-medoids clustering is performed within each of the two scenario subsets based on the comprehensive scenario distance. For scenario subsets with insufficient sample size for clustering, the candidate scenarios within those subsets are directly retained as representative scenarios.

[0215] This embodiment employs the K-medoids clustering method, using the comprehensive scene distance matrix obtained in step S406 as the clustering input. K-medoids clustering selects cluster centers from actual multivariate candidate scenes; therefore, the resulting representative scene remains a complete scene that has passed source-load correlation maintenance verification, operational constraint verification, and joint operational state consistency verification. Compared to using the average of multiple candidate scenes as cluster centers, using actual multivariate candidate scenes as cluster centers avoids the formation of virtual scenes that do not meet the target distribution network operational constraints or lack practical operational meaning due to averaging different scenes.

[0216] First, based on the number of candidate scenarios in the current scenario subset and the scale of scenarios that subsequent security analysis can handle, a lower limit for the number of clusters is set. and the upper limit of the number of clusters Within the following range of cluster sizes:

[0217] In the formula, 2 indicates that when using the silhouette coefficient to evaluate the clustering results, the candidate scene is divided into at least two scene clusters; This indicates the number of candidate scenes contained in the current scene subset being clustered.

[0218] K-medoids clustering was performed on multivariate candidate scenes using different numbers of clusters, and the silhouette coefficients corresponding to each clustering result were calculated. The number of clusters corresponding to the largest silhouette coefficient was determined as the representative number of scenes.

[0219] In the formula, This indicates the number of scenes; Indicates the number of clusters. The silhouette coefficient corresponds to the time. The silhouette coefficient reflects both the tightness between multivariate candidate scenarios within the same candidate scenario cluster and the degree of separation between different candidate scenario clusters. A larger silhouette coefficient indicates greater similarity among multivariate candidate scenarios within the same cluster and more significant differences between different candidate scenario clusters. When multiple cluster sizes correspond to the same silhouette coefficient, the smaller cluster size is selected as the representative number of scenarios to reduce the scale of scenarios that need to be processed in subsequent security analysis. Following the clustering method described above, scenario clusters corresponding to risk candidate scenario subsets and scenario clusters corresponding to non-risk candidate scenario subsets are obtained respectively. All scenario clusters corresponding to risk and non-risk candidate scenario subsets are merged to obtain the candidate scenario cluster set:

[0220] In the formula, Represents the set of candidate scene clusters; Indicates the first A cluster of candidate scenarios; Indicates the cluster index of candidate scenarios; This represents the total number of scenario clusters obtained after clustering the risk candidate scenario subset and the non-risk candidate scenario subset respectively.

[0221] For the For each of the candidate scene clusters, calculate the sum of the comprehensive scene distances between each multivariate candidate scene within the cluster and all other multivariate candidate scenes. Then, determine the actual multivariate candidate scene with the smallest sum of comprehensive scene distances as the representative scene of that candidate scene cluster.

[0222] In the formula, Indicates the first The representative scene corresponding to each candidate scene cluster. When At this time, the comprehensive scene distance between the two is 0, which does not affect the sum of the comprehensive scene distances or the determination result of the representative scene. The representative scene is located at the center of the comprehensive distance of the corresponding candidate scene cluster, and can characterize the main operational characteristics, source-load correlation characteristics, risk characteristics, and time series change characteristics of the multivariate candidate scenes within the candidate scene cluster. According to the first... The number of multivariate candidate scenes contained in each candidate scene cluster is used to calculate the scene probability of the corresponding representative scene:

[0223] In the formula, Indicates the first The number of multivariable candidate scenarios contained in each candidate scenario cluster; Indicates the representative scenario The corresponding scenario probabilities. The scenario probabilities corresponding to each representative scenario satisfy:

[0224] The scenario probability represents the relative proportion of the operating condition represented by the scenario among all multivariate candidate scenarios. The lower the scenario probability, the fewer the number of candidate scenarios contained in the corresponding candidate scenario cluster, and the lower the frequency of the operating condition represented by the scenario in the current scenario generation result.

[0225] This step outputs each representative scenario and its corresponding scenario probability, and inputs them into step S408 for risk attribute and low probability attribute discrimination.

[0226] S408. Based on the risk characteristics and scenario probabilities of representative scenarios, identify low-probability risk scenarios for the distribution network from the representative scenarios. The representative scenarios obtained in step S407 include scenarios with different source load operating conditions, different risk levels, and different probabilities of occurrence. These scenarios may simultaneously include high-probability normal scenarios, low-probability normal scenarios, high-probability risk scenarios, and low-probability risk scenarios. A low probability of a representative scenario alone does not indicate a security access risk; similarly, the presence of risky operational characteristics in a representative scenario alone does not indicate it is a low-probability scenario. Therefore, this step determines whether a representative scenario contains risky operational characteristics and whether its scenario probability meets the low-probability condition.

[0227] S4081. Determine whether the representative scenario contains risky operational characteristics, as follows: According to the risk indicator extraction rules and risk score calculation rules in step S202, the representative scenarios are calculated respectively. Risk score corresponding to each target sampling time When at least one target in the representative scene satisfies the following conditions:

[0228] At that time, determine the representative scenario It includes risk-based operational characteristics. That is, when the following conditions are met:

[0229] At that time, it will represent the scene. This scenario was identified as a representative of the risks.

[0230] It should be noted that step S406 uses the average risk score of each target sampling moment within the target time period to calculate the risk characteristic distance between candidate scenarios. Its purpose is to measure the overall risk level difference between different candidate scenarios throughout the target time period. This step, however, determines whether a target sampling moment that meets the risk discrimination criteria actually exists in the representative scenario. Its purpose is to identify whether the representative scenario contains risky operational characteristics. These two steps are used for scenario distance calculation and risk attribute discrimination, respectively, and do not change the meaning of the risk score of a single target sampling moment in step S202. Using at least one target sampling moment that meets the risk discrimination criteria can prevent short-duration but valuable risky operational states from being diluted by other normal target sampling moments.

[0231] S4082. Determine whether the representative scenario meets the low probability condition, as follows: Representative Scene Corresponding scenario probability Compared with the preset low probability threshold A comparison is made. The preset low-probability threshold is determined based on the range of low-frequency operating scenarios that need to be retained in the target distribution network security access analysis and the historical distribution of the probability of representative scenarios, and is used to distinguish between low-frequency operating conditions and normal operating conditions. The low-probability discrimination condition of the representative scenario is expressed as follows:

[0232] When the scenario probability corresponding to the representative scenario meets the above conditions, the representative scenario is determined as a low-probability representative scenario. The scenario probability reflects the relative proportion of the candidate scenario cluster corresponding to the representative scenario in all multivariate candidate scenarios. Therefore, a scenario probability lower than the preset low-probability threshold indicates that the corresponding operating condition occurs less frequently in the generated scenario set.

[0233] S4083. Identify low-probability risk scenarios for the power distribution network, as detailed below: When the representative scenario simultaneously meets the risk operation characteristic discrimination condition and the low probability discrimination condition, that is:

[0234] and:

[0235] At that time, the representative scenario is identified as a low-probability risk scenario for the distribution network. The final set of low-probability risk scenarios for the distribution network is represented as follows:

[0236] In the formula, This represents a set of low-probability risk scenarios for the power distribution network.

[0237] For each low-probability risk scenario of the distribution network, the following are output: the joint operation status sequence of the target time period, the output sequence of distributed power generation, the load power sequence, the operating power sequence of energy storage and electric vehicles, the node voltage sequence, the load rate sequence of lines and distribution transformers, the risk indication at each target sampling time, the average scenario risk score, and the scenario probability.

[0238] Through the above processing, the final output low-probability risk scenario of the distribution network has the following characteristics: First, the combination of operating variables in the scenario satisfies the operating constraints of the target distribution network and each new entity; Second, the scenario retains the source-load correlation in the historical operating data of the target distribution network; Third, there is at least one target sampling time in the scenario that meets the conditions for safe access risk discrimination; Fourth, the probability of the scenario appearing in all multivariate candidate scenarios is not greater than the preset low-probability threshold.

[0239] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations 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. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A novel method for generating low-probability risk scenarios in distribution networks with primary entity access, characterized in that, include: Based on the historical operation data of the target distribution network, determine the source and load operation status corresponding to the historical operation samples, and identify low-density operation samples; The low-density operation samples that meet the preset data validity conditions and preset risk discrimination conditions are identified as risk operation samples; Based on the source load operating status and the labeling results of the risk operating samples, the joint operating status is determined, a historical joint operating status sequence is formed according to the sampling time, and a Markov state transition model is constructed based on the historical joint operating status sequence. Based on the Markov state transition model, a joint operating state sequence for the target time period is generated, and based on the joint operating state sequence and the corresponding operating samples, a low-probability risk scenario for the distribution network containing risk operation characteristics is generated.

2. The method for generating low-probability risk scenarios in a distribution network with novel subject access as described in claim 1, characterized in that, Determine the source load operating status corresponding to the historical operating samples, including: Based on the distributed power output data and load power data in historical operation samples, the source-load correlation, which characterizes the source-load operation relationship, is extracted. By combining the source-load operation characteristics and source-load correlation of historical operation samples, the historical operation samples are divided into states to obtain the source-load operation status.

3. The method for generating low-probability risk scenarios in a distribution network with novel subject access as described in claim 2, characterized in that, Extracting source-load correlations that characterize the source-load operational relationship, including: Within a preset time window, calculate the source-load correlation coefficient between the output sequence of each distributed power access point and the load power sequence of each load area; Arrange the source-load correlation coefficients according to the correspondence between the distributed power supply access point and the load area, construct the source-load correlation matrix, and determine the source-load correlation based on the source-load correlation matrix.

4. The method for generating low-probability risk scenarios in a distribution network with novel subject access as described in claim 1, characterized in that, Identify low-density running samples, including: Density clustering is performed based on the operational characteristics corresponding to historical operational samples to form operational sample clusters; Historical running samples that are not assigned to any running sample cluster are identified as low-density running samples.

5. The method for generating low-probability risk scenarios in a distribution network with novel subject access as described in claim 4, characterized in that, The low-density operational samples that meet the preset data validity conditions and preset risk discrimination conditions are identified as risky operational samples, including: The validity of low-density operational samples is determined to obtain valid low-density operational samples. Risk indicators are extracted from effective low-density operational samples based on data related to secure access risks, and risk operational samples are determined based on the risk indicators and preset risk discrimination conditions.

6. The method for generating low-probability risk scenarios in a distribution network with novel subject access as described in claim 1, characterized in that, Based on the source-load operating status and the labeling results of the risk operating samples, the joint operating status is determined, and a historical joint operating status sequence is formed according to the sampling time, including: Based on the risk assessment results of each historical operational sample, corresponding risk labels are generated; The source load operation status corresponding to each historical operation sample is combined with the risk marker to obtain the joint operation status of each historical operation sample; The joint operation status is arranged according to the sampling time to form a historical joint operation status sequence, and historical operation samples with the same joint operation status are assigned to the corresponding status sample library.

7. The method for generating low-probability risk scenarios in a distribution network with novel subject access as described in claim 6, characterized in that, Constructing a Markov state transition model includes: Extract the state transition relationships between adjacent joint operation states from the historical joint operation state sequence; Based on the time attributes corresponding to the historical operation samples, the state transition relationships are classified, and the state transition probabilities under different time attributes are calculated. Based on the number of historical transition samples under different time attributes, the state transition probability is corrected using the global state transition probability, and a state transition matrix is ​​constructed based on the corrected state transition probability to form a Markov state transition model.

8. The method for generating low-probability risk scenarios in a distribution network with novel subject access as described in claim 7, characterized in that, Generate a joint operational state sequence for the target time period, including: Obtain the time attributes corresponding to each sampling time within the target time period and match the corresponding state transition matrix; The initial joint operation state of the target time period is taken as the current joint operation state. State transition sampling is performed through the matched state transition matrix to generate the joint operation state sequence of the target time period hour by hour.

9. The method for generating low-probability risk scenarios in a distribution network with novel subject access as described in claim 8, characterized in that, Based on the joint operation state sequence, the corresponding operation sample for the joint operation state is invoked, including: For the joint operation state sequence of the target time period, resampling is performed from the state sample library of the corresponding joint operation state to form the basic scenario; Based on the correlation between the running variables in the corresponding state sample library, the basic scenario is reconstructed by correlation-preserving perturbation. The reconstructed scenario is subjected to runtime constraint verification to obtain multivariate candidate scenarios.

10. The method for generating low-probability risk scenarios in a distribution network with novel subject access as described in claim 9, characterized in that, Generating low-probability risk scenarios for the power distribution network also includes: Based on the operational characteristics, source-load correlation characteristics, and risk characteristics of multivariate candidate scenarios, a comprehensive scenario distance is constructed between multivariate candidate scenarios; Clustering and filtering of multivariate candidate scenes based on comprehensive scene distance yields representative scenes and their corresponding probabilities. Based on the risk characteristics and scenario probabilities of representative scenarios, low-probability risk scenarios for the distribution network are identified from the representative scenarios.

Images