An active distribution network short-circuit parameter online monitoring method, system, device and medium

Abstract

The application discloses an active distribution network short-circuit parameter online monitoring method, system, device and medium, and belongs to the technical field of active distribution network short-circuit parameter monitoring, and comprises the following steps: collecting operation data by applying excitation disturbance to a distributed power supply, constructing an extended phase space state vector through delay embedding, and identifying a current-limiting trigger point from an evolution track; extracting extreme points under multiple working conditions to fit a current-limiting boundary; establishing a state evolution model; iteratively evolving a to-be-measured working condition to a steady state to extract a short-circuit current feature; establishing a mapping relationship between a margin and the short-circuit current feature, and outputting a short-circuit current prediction value and a dominant current-limiting boundary. The application converts multiple physical domain parameters into a state vector, simulates a short-circuit process through iterative evolution, realizes online monitoring and prediction, avoids the risk of equipment damage in an actual short-circuit test, and improves short-circuit current prediction accuracy.

Description

Technical Field

[0001] This invention relates to the field of active distribution network short-circuit parameter monitoring technology, specifically to an online monitoring method, system, device, and medium for active distribution network short-circuit parameters. Background Technology

[0002] With the large-scale grid connection of distributed generation, the distribution network has transformed from a single power source to an active distribution network with multiple power sources. Distributed generation is connected to the grid through power electronic inverters, and its short-circuit current characteristics differ fundamentally from those of traditional synchronous generators. The short-circuit current of the inverter is limited by multiple physical constraints, such as controller saturation, power device thermal protection, and DC bus voltage dips. These constraints belong to the control domain, thermal domain, and energy domain, and are mutually coupled. Traditional methods analyze each physical domain independently, failing to describe the cross-domain coupling mechanism, resulting in large short-circuit current prediction errors. Traditional electromagnetic transient simulation, while capable of describing dynamic processes, is computationally intensive and struggles to meet the real-time requirements of online monitoring. Traditional machine learning methods lack physical interpretability and have weak generalization ability to operating conditions outside the training data. Currently, there is a lack of short-circuit current prediction methods that can accurately describe the multi-domain coupled current-limiting mechanism while also meeting the real-time requirements of online monitoring. Summary of the Invention

[0003] In view of the above-mentioned problems, the present invention provides an active distribution network short-circuit parameter online monitoring method, system, device and medium.

[0004] Therefore, the technical problem solved by this invention is: how to construct a state vector that reflects cross-domain coupling relationship from time series data of electrical parameters in the control domain, electrical parameters in the energy domain, and thermal parameters in the thermal domain; how to identify the constraint boundaries of each physical domain from the current limiting trigger points of multiple operating conditions; how to establish state evolution models under different operating conditions according to the order in which the evolution trajectory crosses the constraint boundaries; and how to quickly query the short-circuit current characteristics and dominant current limiting boundaries of the operating condition under test during online monitoring.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an online monitoring method for short-circuit parameters in an active distribution network, comprising,

[0006] An excitation disturbance is applied to the distributed power source, and the operating data under the excitation disturbance is collected. The operating data is then embedded with a delayed state vector, and the sequence of the state vectors at different times is recorded as an evolution trajectory. The current limiting trigger point is identified from the evolution trajectory.

[0007] The current limiting trigger point is repeatedly identified under multiple operating conditions. Extreme points are extracted from the current limiting trigger points under multiple operating conditions, and the current limiting boundary is obtained by fitting the extreme points.

[0008] Calculate the directed distance from the evolution trajectory to the flow-limiting boundary, classify the evolution trajectory according to the change of the directed distance, and establish a state evolution model for each type of trajectory;

[0009] A short-circuit disturbance is applied to the test condition to obtain an initial state. The state evolution model is selected according to the category to which the initial state belongs. The model is used to iteratively evolve to a steady state, and the short-circuit current characteristics are extracted from the evolution process.

[0010] Calculate the normalized directed distance from the initial state to the current limiting boundary as a margin, establish the mapping relationship between the margin and the short-circuit current characteristics, and output the predicted short-circuit current value and the dominant current limiting boundary.

[0011] As a preferred embodiment of the online monitoring method for short-circuit parameters in an active distribution network according to the present invention, the operating data includes time series of parameters from multiple physical domains;

[0012] The method of constructing a state vector by delaying the embedding of the running data includes: delaying the embedding of each parameter time series in the parameter time series of the multiple physical domains to obtain an embedding vector, and concatenating the embedding vectors to construct the state vector;

[0013] The step of extracting extreme points from the current limiting trigger points of multiple operating conditions includes: extracting a subset of extreme points where each dimension of the embedding vector reaches an extreme value from the current limiting trigger points;

[0014] The process of fitting the extreme points to obtain the current limiting boundary includes: fitting the extreme point subsets to obtain the current limiting boundary, thereby obtaining multiple current limiting boundaries.

[0015] As a preferred embodiment of the online monitoring method for short-circuit parameters in an active distribution network according to the present invention, the parameter time series of the multiple physical domains includes a first electrical parameter time series, a second electrical parameter time series, and a thermal parameter time series.

[0016] A first embedding vector is obtained by delaying the embedding of the first electrical parameter time series, a second embedding vector is obtained by delaying the embedding of the second electrical parameter time series, and a third embedding vector is obtained by delaying the embedding of the thermal parameter time series. The first embedding vector, the second embedding vector, and the third embedding vector are concatenated to construct an extended phase space state vector.

[0017] Extract from the current limiting trigger point the first extreme point set, the second extreme point set, and the third extreme point set in the extended phase space where the first embedded vector dimension reaches an extreme value;

[0018] By fitting the first set of extreme points, the second set of extreme points, and the third set of extreme points respectively, a first current-limiting boundary, a second current-limiting boundary, and a third current-limiting boundary are obtained in the extended phase space. The first current-limiting boundary, the second current-limiting boundary, and the third current-limiting boundary together form a constrained feasible region.

[0019] As a preferred embodiment of the active distribution network short-circuit parameter online monitoring method described in this invention, the first characteristic time scale is used as the delay time when the first electrical parameter time series is delayed and embedded, the second characteristic time scale is used as the delay time when the second electrical parameter time series is delayed and embedded, and the third characteristic time scale is used as the delay time when the thermal parameter time series is delayed and embedded.

[0020] The first characteristic time scale is determined based on the physical response characteristics of the first electrical parameter, the second characteristic time scale is determined based on the physical response characteristics of the second electrical parameter, and the third characteristic time scale is determined based on the physical response characteristics of the thermal parameter.

[0021] As a preferred embodiment of the online monitoring method for short-circuit parameters of an active distribution network according to the present invention, the method involves calculating the first directed distance from each trajectory point on the evolution trajectory to the first current-limiting boundary, the second directed distance to the second current-limiting boundary, and the third directed distance to the third current-limiting boundary.

[0022] Identify the first sign change moment when the first directed distance changes from positive to negative, the second sign change moment when the second directed distance changes from positive to negative, and the third sign change moment when the third directed distance changes from positive to negative. The first sign change moment is the moment when the evolution trajectory crosses the first current-limiting boundary, the second sign change moment is the moment when the evolution trajectory crosses the second current-limiting boundary, and the third sign change moment is the moment when the evolution trajectory crosses the third current-limiting boundary.

[0023] The triggering order of the current limiting boundary is determined based on the chronological order of the first symbol change, the second symbol change, and the third symbol change, and evolution trajectories with the same triggering order are grouped into the same category.

[0024] As a preferred embodiment of the online monitoring method for short-circuit parameters in an active distribution network according to the present invention, the method includes: calculating the directed distance from the initial state to the current limiting boundary, determining the category to which the initial state belongs based on the combination of positive and negative signs of the directed distance, and selecting the state evolution model of that category.

[0025] The state evolution model is used to perform a single-step evolution of the initial state to obtain the state at the next moment;

[0026] Calculate the directed distance from the next time-instance state to the current-limiting boundary;

[0027] When the directional distance sign changes, it is determined that the flow-limiting boundary has been crossed;

[0028] The category is redefined based on the combination of directed distance symbols after the crossing;

[0029] Switch to a new category of state evolution model;

[0030] The state at the next moment is repeatedly subjected to single-step evolution, directed distance calculation, traversal determination, category determination, and model switching for at least one round until the difference between two adjacent evolution states is less than a preset convergence threshold.

[0031] As a preferred embodiment of the online monitoring method for short-circuit parameters in an active distribution network according to the present invention, the step of establishing the mapping relationship between the margin and the short-circuit current characteristics includes: constructing an interpolation function by performing multidimensional interpolation on the margin and the short-circuit current characteristics for multiple operating conditions.

[0032] The output short-circuit current prediction value and dominant current limiting boundary include: collecting operating data during normal operation of the distributed power source; constructing a state vector by delaying the embedding of the operating data; calculating the normalized directed distance from the state vector to the current limiting boundary as a margin; inputting the margin into the interpolation function to obtain the short-circuit current prediction value; calculating the directed distance from the state vector to the current limiting boundary; determining the category to which the state vector belongs based on the combination of positive and negative signs of the directed distance; identifying the first triggered current limiting boundary from the triggering order of the category as the dominant current limiting boundary; and outputting the short-circuit current prediction value and the dominant current limiting boundary.

[0033] This invention provides an online monitoring system for short-circuit parameters in active power distribution networks.

[0034] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an active distribution network short-circuit parameter online monitoring system, comprising:

[0035] The excitation control module is used to control the excitation disturbance applied to the distributed power source;

[0036] The data acquisition module is used to collect operational data under excitation and disturbance conditions.

[0037] The phase space reconstruction module is used to construct state vectors by delaying the embedding of the running data, and record the sequence of state vectors at different times as evolutionary trajectories;

[0038] A current limiting trigger point identification module is used to identify current limiting trigger points from the evolution trajectory;

[0039] The boundary modeling module is used to extract extreme points from the current limiting trigger points of multiple operating conditions and fit the extreme points to obtain the current limiting boundary;

[0040] The trajectory classification module is used to calculate the directed distance from the evolution trajectory to the flow-limiting boundary, classify the evolution trajectory according to the change of the directed distance, and establish a state evolution model for each type of trajectory;

[0041] The short-circuit simulation module is used to apply a short-circuit disturbance to the test condition to obtain an initial state, select the state evolution model according to the category of the initial state, use the model to iteratively evolve to a steady state, and extract short-circuit current characteristics from the evolution process.

[0042] The mapping construction module is used to calculate the normalized directed distance from the initial state to the current limiting boundary as a margin, and to establish a mapping relationship between the margin and the short-circuit current characteristics.

[0043] The online monitoring module is used to output the predicted short-circuit current and the dominant current-limiting boundary.

[0044] The present invention provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the online monitoring method for short-circuit parameters in an active distribution network.

[0045] The present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the active distribution network short-circuit parameter online monitoring method.

[0046] The beneficial effects of this invention are as follows: This invention delays and embeds the time series parameters of different physical domains and splices them to construct an extended phase space state vector, enabling the constraint boundaries of different physical domains to jointly enclose a constrained feasible region in the same phase space, accurately characterizing the cross-domain coupled current limiting mechanism. This invention identifies the moment when the evolution trajectory crosses the current limiting boundary through a change in the sign of the directed distance, determines the triggering order based on the temporal order of the crossing moments, groups trajectories with the same triggering order into the same category, and establishes a state evolution model, revealing the physical current limiting failure modes under different operating conditions. This invention establishes a mapping relationship between operating margin and short-circuit current characteristics offline; during online monitoring, only the normalized directed distance from the real-time state vector to the current limiting boundary needs to be calculated and the mapping relationship queried, avoiding online solution of differential equations. Attached Figure Description

[0047] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 The above is a flowchart of an online monitoring method for short-circuit parameters in an active distribution network, provided as an embodiment of the present invention.

[0049] Figure 2 The flowchart illustrates the current-limiting boundary fitting process of an online monitoring method for short-circuit parameters in an active distribution network, as provided in one embodiment of the present invention.

[0050] Figure 3 The present invention provides a model iteration flowchart for an online monitoring method for short-circuit parameters in an active distribution network according to an embodiment of the present invention. Detailed Implementation

[0051] To make the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0052] Example 1, referring to Figures 1-3 This is one embodiment of the present invention, which provides an online monitoring method for short-circuit parameters in an active distribution network, comprising:

[0053] S1: Apply excitation perturbation to the distributed power source, collect the operating data under the excitation perturbation, construct a state vector by delaying the embedding of the operating data, record the sequence of state vectors at different times as the evolution trajectory, and identify the current limiting trigger point from the evolution trajectory.

[0054] It is understandable that actual short-circuit tests pose a risk of equipment damage. Excitation disturbances drive the inverter output current to increase by increasing the voltage amplitude at the grid connection point during normal inverter operation, thereby obtaining transient response data of the inverter when it approaches the current limiting boundary without triggering a complete short circuit.

[0055] Furthermore, the embedding dimension in the delayed embedding is set to 3. Constraint trigger identification in distributed photovoltaic inverter short-circuit monitoring requires comparing parameter values ​​before and after triggering to determine whether the constraint has been triggered and the magnitude of the trigger. The embedding vector contains data from three time points: parameter values ​​before triggering, parameter values ​​at the moment of triggering, and parameter values ​​after triggering. An embedding dimension of 3 corresponds to these three time points. When the embedding dimension is 2, it is impossible to determine whether the current moment is the moment of triggering or the completion of triggering, and the triggering time positioning error accumulates in the short-circuit current peak prediction error. When the embedding dimension is 4, the fourth time point is located in the normal operation segment before the triggering process. The parameter values ​​in the normal operation segment are small compared to the parameter values ​​before triggering, and the proportion of parameter changes caused by triggering in the embedding vector decreases, leading to a decrease in constraint boundary identification accuracy and an increase in short-circuit current characteristic prediction error. An embedding dimension of 3 is the minimum sufficient dimension for identification in the three stages before, during, and after triggering.

[0056] Furthermore, the current-limiting trigger point is identified through a sudden change in response gain. In active distribution network short-circuit monitoring, the excitation disturbance is an increase in the grid connection point voltage, the parametric response is an increase in the quadrature-axis current output by the inverter, and the response gain is the ratio of the quadrature-axis current increment to the grid connection point voltage increment. During normal operation of the distributed photovoltaic inverter, the quadrature-axis current changes linearly with the grid connection point voltage, and the response gain remains constant. After current-limiting is triggered, the quadrature-axis current is constrained, and the response gain decreases. A current-limiting trigger is determined when the ratio of the response gain to that of two adjacent responses is less than 0.5. The ratio of 0.5 is set based on the inflection point of the short-circuit current waveform corresponding to the constrained trigger point in active distribution network short-circuit monitoring. Before the inflection point, the short-circuit current rises linearly; after the inflection point, the short-circuit current is constrained, and the rate of increase slows down or stops. The response gain ratio reflects the magnitude of the change in the rate of increase of the short-circuit current; a ratio of 0.5 corresponds to a halving of the rate of increase of the short-circuit current. When the rate of increase of the short-circuit current is halved, the impact of the constraint on the short-circuit current is comparable to the impact of the linear upward trend. The constraint begins to significantly affect the short-circuit current waveform. A ratio of less than 0.5 ensures that the constraint corresponding to the identified current-limiting trigger point has significantly affected the short-circuit current waveform, avoiding misjudging the linear upward segment where the constraint has not yet had a significant effect as a trigger, which would cause the predicted short-circuit current peak value to deviate from the actual peak value.

[0057] In some embodiments, a specific implementation of constructing a state vector by delaying the embedding of runtime data in step S1 includes step S11, wherein:

[0058] S11: The running data includes parameter time series of multiple physical domains. Each parameter time series in the parameter time series of multiple physical domains is delayed and embedded to obtain an embedding vector. The embedding vectors are concatenated to construct a state vector.

[0059] It is understandable that the operational data collected in the short-circuit monitoring of distributed photovoltaic inverters includes control domain parameters, energy domain parameters, and thermal domain parameters. The response time of the control domain parameters is determined by the bandwidth of the current loop controller, the response time of the energy domain parameters is determined by the ratio of the DC capacitor value to the power loss, and the response time of the thermal domain parameters is determined by the product of the chip's thermal resistance and thermal capacitance. The controller bandwidth, capacitor time constant, and thermal time constant are physically independent, differing by a factor of 100 from fastest to slowest. During a short circuit in an active distribution network, control constraints, energy constraints, and thermal constraints are triggered sequentially according to the constraint coupling transmission path. The 100-fold difference in response time causes the control constraint triggering process, energy constraint triggering process, and thermal constraint triggering process to unfold sequentially on the time axis. The second constraint triggering process begins only after the first constraint triggering process is completed, and the third constraint triggering process begins only after the second constraint triggering process is completed. The sequential unfolding of the triggering process avoids the inability to distinguish the triggering order due to simultaneous triggering. When triggering simultaneously, the control constraint boundary, energy constraint boundary, and thermal constraint boundary are all crossed at the same time, making it impossible to determine the dominant current limiting boundary and causing the short-circuit current peak prediction to fail. The 100-fold difference in response time ensures that the triggering process unfolds sequentially, making the triggering order identifiable.

[0060] Furthermore, the delay time τi of the i-th physical domain in the delay embedding is taken as the response time constant of that physical domain to a unit step. A unit amplitude step is applied to the i-th physical domain of the distributed photovoltaic inverter, and the response curve of the i-th physical domain parameter is recorded. A first-order inertial element is fitted to the response curve, and the time constant obtained by fitting is the delay time τi. Traditional delay embedding methods use mutual information or autocorrelation function methods to determine the delay time. The delay time determined by the two methods reflects the statistical characteristics of the time series. In active distribution network short-circuit monitoring, the response delay time does not match the physical time window of constraint triggering. The embedding vector cannot simultaneously contain the three time points defined in step S1: the parameter value before triggering, the parameter value at the moment of triggering, and the parameter value after triggering, thus failing to identify constraint triggering. The step response time constant corresponds to the physical response time from never being triggered to being triggered. The embedding dimension 3 in step S1 corresponds to the backtracking history length 2τi. The parameter value defined in step S1 before triggering is located at the third element of the embedding vector corresponding to time t-2τi. The parameter value at the moment of triggering is located at the second element corresponding to time t-τi. The parameter value after triggering is located at the first element corresponding to time t.

[0061] Furthermore, the order of the embedded vectors in the splicing operation is determined based on the constraint coupling transmission path. During a short circuit in a distributed photovoltaic inverter, the constraint coupling transmission path is a cascaded transmission from control to energy to heat. The control domain embedded vectors are arranged at the beginning of the state vector, occupying the first three dimensions; the energy domain embedded vectors are arranged in the middle of the state vector, occupying the middle three dimensions; and the heat domain embedded vectors are arranged at the end of the state vector, occupying the last three dimensions.

[0062] In some embodiments, the specific implementation of the parameter time series of multiple physical domains in step S11 includes step S111, wherein:

[0063] S111: The time series of parameters in multiple physical domains includes a first electrical parameter time series, a second electrical parameter time series, and a thermal parameter time series. The first electrical parameter time series is delayed and embedded to obtain a first embedding vector. The second electrical parameter time series is delayed and embedded to obtain a second embedding vector. The thermal parameter time series is delayed and embedded to obtain a third embedding vector. The first embedding vector, the second embedding vector, and the third embedding vector are concatenated to construct a state vector, which is an extended phase space state vector.

[0064] It is understandable that in the short-circuit monitoring of distributed photovoltaic inverters, the control domain parameters, energy domain parameters, and thermal domain parameters are specifically defined as the first electrical parameter, quadrature-axis current, the second electrical parameter, DC bus voltage, and the thermal parameter, IGBT junction temperature, respectively. In step S11, the arrangement order determined by the constraint coupling transmission path is reflected in step S111 as follows: the first embedded vector, corresponding to the first electrical parameter, quadrature-axis current, is arranged at the beginning of the state vector, occupying the first three dimensions; the second embedded vector, corresponding to the second electrical parameter, DC bus voltage, is arranged in the middle, occupying the middle three dimensions; and the third embedded vector, corresponding to the thermal parameter, IGBT junction temperature, is arranged at the end, occupying the last three dimensions.

[0065] Furthermore, the third, second, and first elements defined in step S11 correspond to the time t-2τ1, t-2τ1, and t-t respectively for the first electrical parameter quadrature-axis current in step S111; the third element for the second electrical parameter DC bus voltage is t-2τ2, t-2τ2, and t-t respectively; and the third element for the thermal parameter IGBT junction temperature is t-2τ3, t-3, and t-t respectively.

[0066] Furthermore, in step S11, the constraint coupling transmission path causes the triggering order to change with the operating conditions, and different triggering orders correspond to different evolution trajectory forms of the state vector in the extended phase space.

[0067] In some embodiments, the specific implementation of the time series delay embedding of the three parameters in step S111 includes step S1111, wherein:

[0068] S1111: When embedding the time series of the first electrical parameter with delay, the first characteristic time scale is used as the delay time; when embedding the time series of the second electrical parameter with delay, the second characteristic time scale is used as the delay time; when embedding the time series of the thermal parameter with delay, the third characteristic time scale is used as the delay time. The first characteristic time scale is determined based on the physical response characteristics of the first electrical parameter, the second characteristic time scale is determined based on the physical response characteristics of the second electrical parameter, and the third characteristic time scale is determined based on the physical response characteristics of the thermal parameter.

[0069] It is understandable that in step S11 of the short-circuit monitoring of distributed photovoltaic inverters, the delay time τi is specified as the first characteristic time scale τ1, the second characteristic time scale τ2, and the third characteristic time scale τ3. The ratio r21 of τ2 to τ1 and the ratio r32 of τ3 to τ2 are both between 50 and 300.

[0070] Furthermore, the lower limit of the ratio r21, 50, corresponds to the minimum proportional relationship between the backtracking history length of the first embedded vector and the backtracking history length of the second embedded vector. The backtracking history length 2τi defined in step S11 indicates that the i-th physical domain embedded vector observes the historical state of the physical domain from time t-2τi to time t at the current time t. In active distribution network short-circuit monitoring, the duration of the constraint triggering process is approximately 3 time constants: the control constraint triggering process lasts approximately 3τ1, the energy constraint triggering process lasts approximately 3τ2, and the thermal constraint triggering process lasts approximately 3τ3. Responding to r21 equaling 50 and τ2 equaling 50τ1, the backtracking history length of the first embedded vector is 2τ1, and the backtracking history length of the second embedded vector, 2τ2, equals 100τ1, meaning the backtracking length of the second embedded vector is 50 times that of the first. In step S11, the 100-fold difference in response time ensures that the energy constraint triggering process begins only after the control constraint triggering process is completed. The control constraint triggering process lasts 3τ1, and the energy constraint triggering process lasts approximately 3τ2, or 150τ1. At the same observation time, the second embedded vector backtracking by 100τ1 can cover the latter two-thirds of the energy constraint triggering process duration of 150τ1, capturing the main process from the start of triggering to its completion. The first embedded vector backtracking by 2τ1 can cover the latter two-thirds of the control constraint triggering process duration of 3τ1, capturing the main stage of the control constraint triggering process. When the ratio is less than 50, the proportion of the second embedded vector backtracking length to the first embedded vector backtracking length is too small, and the second embedded vector backtracking length is less than 100τ1, failing to cover the latter two-thirds of the energy constraint triggering process duration of 150τ1. Incomplete capture of energy constraint triggering information leads to increased prediction error of short-circuit current characteristics. A ratio of 50 is an approximate lower limit for capturing energy constraint triggering information.

[0071] Furthermore, the upper limit of the ratio r21, 300, corresponds to the maximum proportional relationship between the backtracking history length of the first embedded vector and the backtracking history length of the second embedded vector. Responding to r21 equaling 300, τ2 equaling 300τ1, the backtracking history length of the second embedded vector, 2τ2, equals 600τ1, meaning the backtracking length of the second embedded vector is 300 times that of the first embedded vector. The backtracking history length of the first embedded vector is 2τ1, and the backtracking history length of the second embedded vector is 2τ2, or 600τ1, resulting in a ratio of 1:300. In active distribution network short-circuit monitoring, the nine dimensions of the extended phase space state vector should maintain relative numerical balance to avoid some dimensions having too shallow a backtracking depth, leading to insufficient information capture in those dimensions. The first embedded vector occupies the first three dimensions with a backtracking depth of 2τ1, and the second embedded vector occupies the middle three dimensions with a backtracking depth of 600τ1. When the ratio is 1:300, the backtracking depth of the first embedded vector is close to its relative minimum. When the ratio is greater than 300, the backtracking length of the second embedded vector exceeds 600τ1, and the backtracking of the first embedded vector is relatively shallow (2τ1). At the same observation time, the amount of historical information captured by the first embedded vector is too little compared to the second embedded vector. This reduces the weight of the control constraint triggering information in the extended phase space state vector, leading to insufficient capture of control constraint triggering information and increased prediction error of the short-circuit current characteristics. A ratio of 300 is the upper limit to ensure a relatively sufficient backtracking depth for the first embedded vector. The constraint conditions for the ratio r32 are the same as for r21.

[0072] It should be noted that the application of extended phase space state vector in online monitoring of short-circuit parameters of distributed photovoltaic inverters is as follows: when the photovoltaic inverter is running normally in grid connection, the first electrical parameter quadrature axis current, the second electrical parameter DC bus voltage, and the thermal parameter IGBT junction temperature defined in step S111 are collected. A 9-dimensional state vector is constructed by delaying and embedding according to τ1, τ2, and τ3 determined in step S11, and the distance from the state vector to the current limiting boundary fitted in step S2 is calculated.

[0073] S2: Repeatedly identify current limiting trigger points under multiple operating conditions, extract extreme points from the current limiting trigger points under multiple operating conditions, and fit the extreme points to obtain the current limiting boundary.

[0074] Understandably, in the short-circuit monitoring of distributed photovoltaic inverters, multiple operating conditions are obtained by changing the voltage amplitude at the grid connection point and the photovoltaic input power. Following step S1, current-limiting trigger points are identified under these multiple operating conditions. These current-limiting trigger points are distributed across the extended phase space, forming a point set.

[0075] Furthermore, extreme points are extracted from the current-limiting trigger points under multiple operating conditions. Extreme points are those where the coordinates in each dimension reach their maximum or minimum values. In active distribution network short-circuit monitoring, extreme points correspond to the critical operating states of the inverter on each constraint boundary. In the critical operating state, the constraint is just triggered, and the inverter's output capacity reaches the limit of that constraint. Non-extreme points correspond to operating states where the constraint has been triggered but there is a margin; in the operating state where the constraint has been triggered, the inverter's output capacity is less than the constraint limit. The first, second, and third current-limiting boundaries describe the critical states where control constraints, energy constraints, and thermal constraints are just triggered, respectively. Extreme points are located on the current-limiting boundaries, while non-extreme points are located inside the current-limiting boundaries. Extreme points are extracted to fit the current-limiting boundaries; non-extreme points are not included in the fitting to avoid deviations from the true boundary positioning, which could increase the prediction error of the short-circuit current peak.

[0076] In some embodiments, the specific implementation of extracting extreme points from the current limiting trigger points of multiple operating conditions and fitting the current limiting boundary in step S2 includes steps S21-S22, wherein:

[0077] S21: Extract a subset of extreme points where each dimension of the embedded vector reaches an extreme value from the rate limiting trigger point.

[0078] It is understandable that the state vector constructed in step S1 is a 9-dimensional vector, with the 9 dimensions divided into three groups. The first 3 dimensions correspond to the first electrical parameter, quadrature-axis current, first embedded vector, as defined in step S111; the middle 3 dimensions correspond to the second electrical parameter, DC bus voltage, second embedded vector; and the last 3 dimensions correspond to the thermal parameter, IGBT junction temperature, third embedded vector. Reaching an extreme value for each dimension means that the coordinate of the i-th dimension reaches its maximum or minimum value among all current-limiting trigger points. The extreme value points of the first 3 dimensions that reach their extreme values ​​are extracted from the current-limiting trigger points to form the first extreme value subset; the extreme value points of the middle 3 dimensions that reach their extreme values ​​are extracted to form the second extreme value subset; and the extreme value points of the last 3 dimensions that reach their extreme values ​​are extracted to form the third extreme value subset.

[0079] S22: Fit the extreme point subsets to obtain the current limiting boundary, resulting in multiple current limiting boundaries.

[0080] Understandably, fitting the first extreme point subset yields the first current-limiting boundary, fitting the second extreme point subset yields the second current-limiting boundary, and fitting the third extreme point subset yields the third current-limiting boundary. The fitting process ensures that the first, second, and third current-limiting boundaries are independent of each other, with the first current-limiting boundary determined solely by the first extreme point subset and unaffected by the second and third extreme point subsets. Similarly, in active distribution network short-circuit monitoring, the triggering conditions for control constraints, energy constraints, and thermal constraints are independent. Fitting the first, second, and third current-limiting boundaries ensures that their independence corresponds to the control constraint boundary, the second current-limiting boundary corresponds to the energy constraint boundary, and the third current-limiting boundary corresponds to the thermal constraint boundary.

[0081] In some embodiments, the specific implementation of extracting a subset of extreme points and fitting multiple current-limiting boundaries in steps S21-S22 includes step S211, wherein:

[0082] S211: Extract the set of first extreme points, the set of second extreme points, and the set of third extreme points in the extended phase space that reach extreme values ​​in the first embedding vector dimension from the current limiting trigger point. Fit the first extreme point set, the second extreme point set, and the third extreme point set to obtain the separating hyperplane. Use the least squares method to determine the hyperplane parameters to minimize the sum of squared distances from the extreme points to the hyperplane. Obtain the first current limiting boundary, the second current limiting boundary, and the third current limiting boundary in the extended phase space. The first current limiting boundary, the second current limiting boundary, and the third current limiting boundary together form the constrained feasible region.

[0083] It is understandable that the first subset of extreme points defined in step S21 is the first set of extreme points in step S211, and the points in the first set of extreme points reach their extreme values ​​in the first three dimensions defined in step S111, namely the first embedded vector dimension of the first electrical parameter, cross-axis current. The second subset of extreme points defined in step S21 is the second set of extreme points in step S211, and the points in the second set of extreme points reach their extreme values ​​in the middle three dimensions defined in step S111, namely the second embedded vector dimension of the second electrical parameter, DC bus voltage. The third subset of extreme points defined in step S21 is the third set of extreme points in step S211, and the points in the third set of extreme points reach their extreme values ​​in the last three dimensions defined in step S111, namely the third embedded vector dimension of the thermal parameter, IGBT junction temperature.

[0084] Furthermore, the first, second, and third current-limiting boundaries defined in step S22 correspond to the control constraint boundary, energy constraint boundary, and thermal constraint boundary in step S211. The constrained feasible region is the area in the extended phase space that simultaneously satisfies the control constraint, energy constraint, and thermal constraint. The state vector within the region corresponds to the inverter's operating state when no constraints are triggered, while the state vector outside the region corresponds to the inverter's operating state when at least one constraint is triggered. In active distribution network short-circuit monitoring, the inverter's normal operating state is located within the constrained feasible region. During a short-circuit fault, the state vector evolves from within the feasible region to the feasible region boundary. Crossing the boundary corresponds to constraint triggering, and the order in which the boundaries are crossed corresponds to the constraint triggering order. Different triggering orders correspond to different short-circuit current waveforms.

[0085] S3: Calculate the directed distance from the evolution trajectory to the flow-limiting boundary, classify the evolution trajectory according to the change of the directed distance, and establish a state evolution model for each type of trajectory.

[0086] It is understandable that the evolution trajectory recorded in step S1 is a time series of the state vector during the application of the excitation disturbance. The directed distance is the distance from the state vector to the current-limiting boundary fitted in step S2. The distance has a positive or negative sign; a positive value indicates that the state vector is within the feasible region of the constraints defined in step S211, and a negative value indicates that the state vector is outside the feasible region of the constraints. In active distribution network short-circuit monitoring, the directed distance reflects the margin between the current operating state of the inverter and the constraint triggering; a positive value corresponds to the constraint not being triggered, and a negative value corresponds to the constraint being triggered.

[0087] Furthermore, the evolutionary trajectories are classified based on the change in directed distance. As the evolutionary trajectory moves from within the feasible region to its boundary in the extended phase space, the directed distance changes from a positive value to a negative value. The moment the directed distance changes sign corresponds to the moment the evolutionary trajectory crosses the current-limiting boundary. The three current-limiting boundaries defined in step S211 correspond to the three constraint boundaries, and the order in which the evolutionary trajectory crosses these boundaries corresponds to the triggering order caused by the coupling propagation path in step S11. Different triggering orders classify the evolutionary trajectories into different categories.

[0088] Furthermore, a state evolution model is established for each type of trajectory. The state evolution model describes the change of the state vector over time, and the evolution law is determined by the dynamic characteristics of the inverter under the corresponding constraint combination. Evolutionary trajectories of the same category have the same triggering order; when the triggering order is the same, the constraint combination is the same; when the constraint combination is the same, the dynamic characteristics are the same, and the evolution law is the same. In active distribution network short-circuit monitoring, different categories correspond to different constraint triggering states, and the evolution law of each category is different. Therefore, an independent grouped convolutional neural network is trained for each category as the state evolution model for that category.

[0089] The state evolution model is established using a grouped convolutional neural network. The neural network structure design reflects the physical structure of the extended phase space state vector defined in step S111. The neural network consists of three layers: a grouped feature extraction layer, a feature fusion layer, and a state prediction layer.

[0090] The group feature extraction layer extracts features from the three sets of embedding vectors of the 9-dimensional state vector constructed in step S1. The 9 dimensions are divided into three groups: the first 3 dimensions correspond to the first electrical parameter, the cross-axis current, the first embedding vector defined in step S111; the middle 3 dimensions correspond to the second electrical parameter, the DC bus voltage, the second embedding vector; and the last 3 dimensions correspond to the thermal parameter, the IGBT junction temperature, the third embedding vector. The first feature extraction sub-network processes the first embedding vector, including a one-dimensional convolutional layer and a fully connected layer. The one-dimensional convolutional layer has a kernel size of 3, corresponding to the embedding dimension of 3 defined in step S1. The number of convolutional kernels is 12, corresponding to an embedding dimension of 3 expanded by 4 times. Each embedding vector contains 3 time points. The 12 convolutional kernels fully extract features from the combination patterns of the 3 time points. The 4-fold expansion extracts 4 types of feature patterns: individual features before triggering, individual features at the moment of triggering, individual features after triggering, and the combination of the three. The convolution operation extracts local features from the parameter values ​​at the three time points of the first embedding vector: parameter values ​​before triggering, parameter values ​​at the moment of triggering, and parameter values ​​after triggering. The fully connected layer maps the 12-dimensional features output by the convolutional layer into a 9-dimensional feature vector. The 9-dimensional feature vector represents the dynamic characteristics of the control domain in the three stages of before triggering, at the moment of triggering, and after triggering. Three key feature parameters are extracted in each stage. The 9-dimensional feature vector corresponds to the 9-dimensional state vector constructed in step S1, which facilitates subsequent feature fusion. The second feature extraction sub-network processes the second embedding vector, with the same structure as the first feature extraction sub-network. The fully connected layer outputs a 9-dimensional feature vector representing the dynamic characteristics of the energy domain. The third feature extraction sub-network processes the third embedding vector, with the same structure as the first feature extraction sub-network. The fully connected layer outputs a 9-dimensional feature vector representing the dynamic characteristics of the thermal domain. This grouped feature extraction reflects the physical independence of the control domain parameters, energy domain parameters, and thermal domain parameters in step S11, with each of the three sub-networks independently processing one of the three physical domains.

[0091] The feature fusion layer fuses the features extracted from the three sub-networks. The input is a 27-dimensional feature vector obtained by concatenating three 9-dimensional feature vectors. It includes a first fully connected layer and a second fully connected layer. The first fully connected layer has 54 neurons, corresponding to a 27-dimensional feature vector that is expanded by a factor of 2 to facilitate learning the complex interactions between features. The activation function used is the Rectified Linear Unit (ReLU) function. The second fully connected layer has 27 neurons, corresponding to compressing the output of the first fully connected layer back to 27 dimensions to retain key features. The activation function used is also the ReLU function. The feature fusion layer learns the coupling relationships between the three physical domains caused by the constraint coupling propagation path defined in step S11. The fused 27-dimensional feature vector represents the overall dynamic characteristics under the coupling effects of the control domain, energy domain, and thermal domain.

[0092] The state prediction layer predicts the state vector for the next time step based on the fused features. It consists of three sub-layers: a first predictive fully connected layer, a second predictive fully connected layer, and an output layer. The first predictive fully connected layer has 54 neurons, corresponding to a 27-dimensional fused feature expanded by a factor of 2 to provide sufficient predictive power. The ReLU activation function is used. The second predictive fully connected layer also has 27 neurons, compressing the output of the first predictive fully connected layer to the same dimension as the fused features. The ReLU activation function is also used. The output layer has 9 neurons, corresponding to the 9-dimensional state vector for the next time step. A linear activation function is used to ensure that the output value can be any real number.

[0093] In active distribution network short-circuit monitoring, the neural network input is the current state vector, which includes the current value of the first electrical parameter (quasi-axis current), the value of the first electrical parameter (quasi-axis current) one time delay, the value of the first electrical parameter (quasi-axis current) two time delays ago, the current value of the second electrical parameter (DC bus voltage), the value of the second electrical parameter (DC bus voltage) one time delay, the value of the second electrical parameter (DC bus voltage) two time delays ago, the current value of the thermal parameter (IGBT junction temperature), the value of the thermal parameter (IGBT junction temperature) one time delay, and the value of the thermal parameter (IGBT junction temperature) two time delays ago. The delay times are the first characteristic time scale τ1, the second characteristic time scale τ2, and the third characteristic time scale τ3 determined in step S11. The neural network output is the state vector for the next time step, which has the same structure as the input but is advanced by one time step. The input state vector reflects the current operating state and historical state of the inverter, while the output state vector reflects the next operating state and historical state. The neural network learns the mapping relationship from the current state to the next state, reflecting the dynamic characteristics of the inverter under the corresponding constraint-triggered state. The time step is determined according to the first feature time scale τ1 defined in step S11, and is set to 0.1 times τ1. Since τ1 is the response time constant of the control domain parameter, setting the time step to 0.1 times τ1 ensures that the time step is less than the control domain response time, guaranteeing that the neural network can capture the dynamic changes of the control constraint triggering process. The control constraint triggering process lasts approximately 3τ1, and the time step of 0.1τ1 results in 30 time steps. These 30 time steps satisfy the sampling theorem requirement, and a sampling frequency of more than 10 times the signal frequency can sufficiently capture signal features. 30 time steps are sufficient to capture the dynamic changes of the triggering process with sufficient precision. Increasing the number of time steps would increase computation but would only slightly improve prediction accuracy.

[0094] Furthermore, the neural network training process includes six steps: training data preparation, network initialization, forward propagation, loss calculation, backpropagation, and parameter update.

[0095] Training data preparation involves extracting trajectory segments belonging to the same category from the evolution trajectories of multiple operating conditions in step S2. These trajectory segments share the same constraint triggering state. The state vectors of adjacent time steps within a trajectory segment constitute training sample pairs. Each training sample contains an input state vector and an output state vector. The input state vector is the state vector at a specific time step, and the output state vector is the state vector one time step backward from that time step. Responding to the multiple operating conditions in step S2, the training data is obtained by changing the grid connection point voltage amplitude and photovoltaic input power. This training data covers the evolution trajectories under different voltage amplitudes and power levels, and the broad coverage enables the trained neural network to have generalization capabilities.

[0096] The network initialization uses the Xavier initialization method for the convolutional kernels of convolutional layers and the connection weights of fully connected layers, with the bias terms initialized to 0. The Xavier initialization method determines the range of initial weight values ​​based on the number of input and output neurons, ensuring that the output variance of each layer remains stable during forward propagation and avoiding gradient vanishing or gradient exploding.

[0097] The forward propagation computes the neural network output. The input state vector is divided into three groups: the first three elements are input to the first feature extraction sub-network, the middle three elements to the second feature extraction sub-network, and the last three elements to the third feature extraction sub-network. The first feature extraction sub-network's convolutional layer contains 12 convolutional kernels, each with a size of 3. It performs one-dimensional convolution on the three input elements, with each kernel outputting one feature value. The 12 kernels output 12-dimensional features. The convolutional layer uses the ReLU activation function, and the fully connected layer maps the 12-dimensional features to a 9-dimensional feature vector. The second and third feature extraction sub-networks perform the same computation, each outputting a 9-dimensional feature vector. The feature fusion layer concatenates the three 9-dimensional feature vectors into a 27-dimensional feature vector, expands it to 54 dimensions through the first fully connected layer, and then compresses it back to 27 dimensions through the second fully connected layer. The state prediction layer expands the 27-dimensional fused features to 54 dimensions through the first prediction fully connected layer, compresses it back to 27 dimensions through the second prediction fully connected layer, and finally maps it to a 9-dimensional predicted state vector through the output layer.

[0098] The loss calculation uses a weighted mean square error loss function, which calculates the error between the predicted state vector and the actual state vector. The error is divided into three groups for calculation: the first group is the mean square error between the first three elements of the predicted state vector and the first three elements of the actual state vector, corresponding to the prediction error of the first electrical parameter, quadrature-axis current; the second group is the mean square error between the middle three elements of the predicted state vector and the middle three elements of the actual state vector, corresponding to the prediction error of the second electrical parameter, DC bus voltage; and the third group is the mean square error between the last three elements of the predicted state vector and the last three elements of the actual state vector, corresponding to the prediction error of the thermal parameter, IGBT junction temperature. The total loss is obtained by weighted summation of the three groups of errors. The weights are determined based on the 100-fold difference in response time in step S11; the faster the response time, the greater the impact on the short-circuit current peak, and the higher the weight. The control domain response time is τ1, with a corresponding weight of 1.0. The energy domain response time is approximately 100τ1, with a corresponding weight of 0.1. The thermal domain response time is approximately 10000τ1, with a corresponding weight of 0.01. The weight ratio 1:0.1:0.01 adopts the reciprocal of the square root of the response time ratio 1:100:10000. A weight of 0.1 corresponds to the reciprocal of the square root of 100, and a weight of 0.01 corresponds to the reciprocal of the square root of 10000. This avoids excessive weight differences caused by direct reciprocals. Longer response times result in smaller weights, maintaining a reasonable weight distribution. The weighted loss function prioritizes learning the dynamic characteristics of the control domain with faster responses, ensuring the accuracy of short-circuit current peak prediction. Since the short-circuit current peak is mainly determined by the triggering time of control constraints, the prediction accuracy of the control domain dynamic characteristics directly affects the prediction accuracy of the short-circuit current peak.

[0099] Backpropagation calculates the gradient of the loss function with respect to the weights and biases of each layer, using a chain rule to calculate the gradient layer by layer from the output layer to the input layer. Parameter updates employ an Adaptive Moment Estimation (Adam) optimizer to update the weights and biases, with a learning rate of 0.001 and 2000 training epochs. Each epoch uses all training samples for forward propagation, loss calculation, backpropagation, and parameter updates. The loss function value decreases as the number of training epochs increases, and training stops when the loss function value converges to a preset threshold or when the number of training epochs reaches 2000.

[0100] In active distribution network short circuit monitoring, the trained neural network state evolution model is used to predict the evolution trajectory of the state vector during short circuit faults in step S4. The predicted evolution trajectory provides the time-domain characteristics of the short circuit current waveform.

[0101] In some embodiments, the specific implementation of classifying evolutionary trajectories based on directional distance changes in step S3 includes steps S31-S32, wherein:

[0102] S31: Calculate the first directed distance from each trajectory point on the evolution trajectory to the first flow-limiting boundary, the second directed distance to the second flow-limiting boundary, and the third directed distance to the third flow-limiting boundary.

[0103] It is understandable that in step S1, the evolution trajectory is a sequence of state vectors at different times, and each trajectory point on the evolution trajectory is a state vector in the sequence. The k-th trajectory point is the state vector at the k-th time, and the k-th trajectory point is the 9-dimensional vector constructed in step S1. The first directed distance is the distance from the k-th trajectory point to the first current-limiting boundary defined in step S211, the second directed distance is the distance from the k-th trajectory point to the second current-limiting boundary, and the third directed distance is the distance from the k-th trajectory point to the third current-limiting boundary.

[0104] Furthermore, the positional relationship between the evolution trajectory and the three current-limiting boundaries is quantified by calculating three directed distances. In step S211, the three current-limiting boundaries correspond to the three constraints defined in step S111: the first current-limiting boundary corresponds to the control constraint boundary, the second current-limiting boundary corresponds to the energy constraint boundary, and the third current-limiting boundary corresponds to the thermal constraint boundary. In active distribution network short-circuit monitoring, the three directed distances reflect the margin between the inverter's current operating state and the three constraints. The sign combination of the three directed distances determines the constraint triggering state. The sign combination of the three directed distances is denoted as (symbol 1, symbol 2, symbol 3), where symbol 1 is the symbol of the first directed distance, symbol 2 is the symbol of the second directed distance, and symbol 3 is the symbol of the third directed distance. Positive values ​​are represented by +, and negative values ​​by -. The symbol combination (+,+,+) indicates that the control constraint, energy constraint, and thermal constraint are not triggered; the symbol combination (-,+,+) indicates that the control constraint is triggered but the energy constraint and thermal constraint are not triggered; the symbol combination (-,-,+) indicates that the control constraint and energy constraint are triggered but the thermal constraint is not triggered; and the symbol combination (-,-,-) indicates that the control constraint, energy constraint, and thermal constraint are all triggered.

[0105] S32: Identify the first sign change moment when the first directed distance changes from positive to negative, the second sign change moment when the second directed distance changes from positive to negative, and the third sign change moment when the third directed distance changes from positive to negative. The first sign change moment is the moment when the evolutionary trajectory crosses the first current-limiting boundary, the second sign change moment is the moment when the evolutionary trajectory crosses the second current-limiting boundary, and the third sign change moment is the moment when the evolutionary trajectory crosses the third current-limiting boundary. Determine the triggering order of the current-limiting boundaries based on the chronological order of the first, second, and third sign change moments, and group evolutionary trajectories with the same triggering order into the same category.

[0106] It is understandable that the change of the directed distance from positive to negative corresponds to the evolution trajectory crossing the boundary from within the feasible region defined in step S211 to outside the feasible region, and the boundary crossing corresponds to the constraint changing from never being triggered to being triggered. The first sign change of the first directed distance from positive to negative corresponds to the control constraint triggering time, the second sign change of the second directed distance from positive to negative corresponds to the energy constraint triggering time, and the third sign change of the third directed distance from positive to negative corresponds to the thermal constraint triggering time.

[0107] Furthermore, the triggering order of the current-limiting boundary is determined based on the temporal order of the three symbol changes. When the first symbol change occurs before the second symbol change occurs before the third symbol change, the triggering order is from control to energy to heat; when the third symbol change occurs before the first symbol change occurs before the second symbol change, the triggering order is from heat to control to energy. The triggering order corresponds to the different transmission paths of the coupling transmission path in step S11.

[0108] Furthermore, evolution trajectories with the same triggering order are grouped into the same category. Evolution trajectories of the same category cross the three current-limiting boundaries in the same order, and when the crossing order is the same, the constraint triggering order is also the same. In active distribution network short-circuit monitoring, the constraint triggering order determines the short-circuit current waveform. When the control constraint triggers first, the short-circuit current reaches its peak at control saturation; when the thermal constraint triggers first, the short-circuit current reaches its peak at thermal protection. Evolution trajectories of the same category correspond to the same short-circuit current waveform characteristics, while different categories correspond to different short-circuit current waveform characteristics. The state evolution model established for each type of trajectory reflects the dynamic characteristics of the corresponding category, and the model is used to predict the short-circuit current waveform belonging to that category.

[0109] S4: Apply a short-circuit disturbance to the test condition to obtain the initial state, select the state evolution model according to the category of the initial state, use the model to iteratively evolve to the steady state, and extract the short-circuit current characteristics from the evolution process.

[0110] It is understandable that the operating condition under test in active distribution network short-circuit monitoring is the normal grid-connected operation of distributed photovoltaic inverters, and the short-circuit disturbance is achieved by short-circuiting the three phases at the grid connection point. The difference between the short-circuit disturbance and the excitation disturbance in step S1 is that the excitation disturbance applies an increased voltage amplitude during normal operation, while the short-circuit disturbance applies a three-phase short circuit under the operating condition under test. At the instant the short-circuit disturbance is applied, the first electrical parameter quadrature-axis current, the second electrical parameter DC bus voltage, and the thermal parameter IGBT junction temperature defined in step S111 are collected, and a 9-dimensional state vector is constructed as the initial state according to the delays of τ1, τ2, and τ3 determined in step S11.

[0111] Furthermore, the state evolution model is selected from the multiple evolution models established in step S3 based on the category to which the initial state belongs.

[0112] Furthermore, iterative evolution to steady state refers to repeatedly using the state evolution model to calculate the state at the next moment until the state no longer changes. During a short circuit in an active distribution network, the inverter evolves from the initial short-circuit state to the short-circuit steady state. During this evolution, the state vector moves in the extended phase space, and the corresponding constraint is triggered when the trajectory crosses the current-limiting boundary identified in step S2. Iterative evolution simulates the evolution trajectory of the state vector during the short circuit, and the order in which the evolution trajectory crosses the current-limiting boundary is the constraint triggering order.

[0113] It should be noted that the short-circuit current characteristics include the peak value and rise time. The peak value is the maximum value reached by the first electrical parameter, the quadrature-axis current, during the iterative evolution process, while the rise time is the time from the initial state to the first electrical parameter, the quadrature-axis current, reaching its peak value. The peak value and rise time are determined by the constraint triggering sequence; different triggering sequences correspond to different peak values ​​and rise times.

[0114] In some embodiments, the specific implementation of step S4, which uses model iterative evolution to a steady state, includes steps S41-S47, wherein:

[0115] S41: Calculate the directed distance from the initial state to the flow-limiting boundary, determine the category of the initial state based on the combination of positive and negative signs of the directed distance, and select the state evolution model of that category.

[0116] It is understandable that the directed distance is the distance from the state vector to the current-limiting boundary identified in step S2. The directed distance is positive when the state vector is outside the current-limiting boundary and negative when it is inside the current-limiting boundary. The directed distances from the initial state in step S4 to the first, second, and third current-limiting boundaries defined in step S211 are d1, d2, and d3, respectively.

[0117] Furthermore, the positive and negative sign combinations are the sign combinations d1, d2, and d3. In active distribution network short-circuit monitoring, the first, second, and third current-limiting boundaries divide the extended phase space into multiple regions. The directed distance sign combination from the state vector within each region to the first, second, and third current-limiting boundaries is the same. The sign combinations (+,+,+), (-,+,+), (-,-,+), and (-,-,-) defined in step S31 represent different constraint triggering states. The sign combination of the initial state determines the category to which the initial state belongs. The state evolution model established for each category in step S3 is selected according to the category in step S41.

[0118] S42: Use a state evolution model to perform a single-step evolution of the initial state to obtain the state at the next moment.

[0119] S43: Calculate the directed distance from the state at the next time step to the flow-limiting boundary.

[0120] It is understandable that the directed distances from the next time-instance state obtained in step S42 to the first current-limiting boundary, the second current-limiting boundary, and the third current-limiting boundary defined in step S211 are d1', d2', and d3', respectively.

[0121] S44: Determine if the direction distance sign changes when crossing the flow-limiting boundary.

[0122] Understandably, in response to the different signs of d1 and d1', it is determined that the first current-limiting boundary has been crossed, corresponding to the control constraint triggering defined in step S211. In response to the different signs of d2 and d2', it is determined that the second current-limiting boundary has been crossed, corresponding to the energy constraint triggering. In response to the different signs of d3 and d3', it is determined that the third current-limiting boundary has been crossed, corresponding to the thermal constraint triggering.

[0123] S45: Re-determine the category based on the combination of directed distance symbols after crossing.

[0124] S46: Switch to the state evolution model of the new category.

[0125] It is understandable that the new category determined in step S45 corresponds to the state evolution model established in step S3.

[0126] S47: Repeat the single-step evolution, directed distance calculation, traversal determination, category determination, and model switching for the next time step state at least once, until the difference between two adjacent evolution states is less than the preset convergence threshold.

[0127] It is understandable that during a short circuit in an active distribution network, the inverter gradually evolves from its initial state to a steady state.

[0128] Furthermore, the preset convergence threshold is the difference threshold between two adjacent evolution states. In active distribution network short-circuit monitoring, after the short-circuit steady-state constraint is triggered, the inverter output remains constant, and the state vector no longer changes. The difference between two adjacent evolution states is the Euclidean distance in the extended phase space between the state vector obtained in the nth iteration and the state vector obtained in the (n-1)th iteration. The Euclidean distance is calculated as the square root of the sum of the squares of the differences in the nine coordinate dimensions. The convergence threshold is determined based on the short-circuit current prediction accuracy requirements. In active distribution network short-circuit monitoring, the coordinate range of each dimension of the state vector reflects the range of changes in the inverter's operating state, and the peak short-circuit current is proportional to the magnitude of the state vector coordinate change. The proportion of the difference between two adjacent state vector evolutions to the coordinate range reflects the relative magnitude of the short-circuit current peak change; when the proportion of the difference equals the prediction accuracy requirement, it corresponds to the convergence criterion. The convergence threshold is set to 0.05 times the coordinate range of each dimension of the initial state in step S4 when the peak relative error of the prediction accuracy is less than 5%. It is set to 0.02 times the coordinate range of each dimension of the initial state when the peak relative error of the prediction accuracy is less than 2%, and to 0.01 times the coordinate range of each dimension of the initial state when the peak relative error of the prediction accuracy is less than 1%. A smaller threshold results in more iterations and longer computation time but higher prediction accuracy; conversely, a larger threshold results in fewer iterations, shorter computation time, but lower prediction accuracy.

[0129] S5: Calculate the normalized directed distance from the initial state to the current limiting boundary as a margin, establish the mapping relationship between the margin and the short-circuit current characteristics, and output the predicted short-circuit current value and the dominant current limiting boundary.

[0130] Understandably, the normalized directed distance is the directed distance calculated in step S4 divided by the range of coordinates of each dimension of the state vector in the extended phase space. In active distribution network short-circuit monitoring, the extended phase space is a 9-dimensional space. These 9 dimensions correspond to the three dimensions of the first electrical parameter (cross-axis current, first embedded vector), the three dimensions of the second electrical parameter (DC bus voltage, second embedded vector), and the three dimensions of the thermal parameter (IGBT junction temperature, third embedded vector) defined in step S111. The physical dimensions of each dimension are different: the first electrical parameter (cross-axis current) dimension is in current units, the second electrical parameter (DC bus voltage) dimension is in voltage units, and the thermal parameter (IGBT junction temperature) dimension is in temperature units. Coordinates with different physical dimensions cannot be directly compared. Normalization converts the directed distances with different physical dimensions into dimensionless relative distances, which can be directly compared. The normalized directed distance is the ratio of the distance from the state vector to the current-limiting boundary to the range of the state vector coordinates. This ratio reflects the relative distance between the state vector and the current-limiting boundary.

[0131] Furthermore, the margin is a normalized directed distance. In active distribution network short-circuit monitoring, the margin reflects how close the inverter's current operating state is to the current limiting trigger. A large margin indicates that the inverter is far from the current limiting trigger and has a large operating safety margin, while a small margin indicates that the inverter is close to the current limiting trigger and has a small operating safety margin. The normalized directed distances from the initial state in step S4 to the first current limiting boundary, the second current limiting boundary, and the third current limiting boundary defined in step S211 are m1, m2, and m3, respectively.

[0132] Furthermore, the mapping relationship is a functional relationship between the margin and the short-circuit current characteristics defined in step S4. In active distribution network short-circuit monitoring, the margin of the inverter's current operating state determines the peak value and rise time of the short-circuit current during a short circuit. A larger margin results in a smaller peak value and a longer rise time for the short-circuit current, while a smaller margin results in a larger peak value and a shorter rise time for the short-circuit current. The mapping relationship is established using margin and short-circuit current characteristic data from multiple operating conditions. The input to the mapping relationship is the margin of the current operating state, and the output is the predicted peak value and rise time of the short-circuit current.

[0133] It should be noted that the dominant current-limiting boundary is the first current-limiting boundary that is crossed during the short circuit process among the first, second, and third current-limiting boundaries defined in step S211. In active distribution network short-circuit monitoring, the first current-limiting boundary crossed corresponds to the first triggered constraint. The first triggered constraint dominates the inflection point and peak value of the short-circuit current waveform. Once the dominant current-limiting boundary is determined, it is possible to identify which constraint limits the short-circuit current.

[0134] In some embodiments, the specific implementation of establishing and outputting the mapping relationship between the margin and the short-circuit current characteristics in step S5 includes steps S51-S52, wherein:

[0135] S51: Construct an interpolation function by performing multidimensional interpolation on the margin and short-circuit current characteristics of multiple operating conditions.

[0136] It is understandable that the multiple operating conditions are those obtained in step S2 by changing the voltage amplitude at the grid connection point and the photovoltaic input power. Under the k-th operating condition, a short-circuit disturbance is applied to the operating condition under test as in step S4. The normalized directed distances m1k, m2k, and m3k from the initial state defined in step S4 to the first, second, and third current-limiting boundaries are calculated as margins. The peak short-circuit current Ipeakk and rise time trk are extracted from step S4 as short-circuit current features. The data from multiple operating conditions constitute a dataset, where the k-th data point is (m1k, m2k, m3k, Ipeakk, trk).

[0137] Furthermore, in multidimensional interpolation, multidimensional means that the input of the interpolation function is a 3-dimensional vector consisting of three margins m1, m2, and m3, and the output is a 2-dimensional vector consisting of two short-circuit current characteristics Ipeak and tr.

[0138] S52: Collect operating data during normal operation of the distributed power source, construct a state vector by delaying the embedding of the operating data, calculate the normalized directed distance from the state vector to the current limiting boundary as a margin, input the margin into the interpolation function to obtain the short-circuit current prediction value, calculate the directed distance from the state vector to the current limiting boundary, determine the category to which the state vector belongs based on the combination of positive and negative signs of the directed distance, identify the first triggered current limiting boundary from the triggering order of the category as the dominant current limiting boundary, and output the short-circuit current prediction value and the dominant current limiting boundary.

[0139] Understandably, the operational data during normal operation differs from the data acquisition time at the moment the short-circuit disturbance is applied in step S4. The data acquired at the moment the short-circuit disturbance is applied in step S4 constructs the initial state for iterative evolution to simulate the short-circuit process, while the data acquired during normal operation in step S52 constructs a state vector for online monitoring and prediction of the short-circuit current. During normal operation, without short-circuit disturbance, the first electrical parameter (quasi-axis current), the second electrical parameter (DC bus voltage), and the thermal parameter (IGBT junction temperature) defined in step S111 are acquired, and a 9-dimensional state vector is constructed using the same delayed embedding method as in step S4.

[0140] Furthermore, the directed distances d1, d2, and d3 from the state vector to the first, second, and third current-limiting boundaries defined in step S211 are calculated and normalized to obtain m1, m2, and m3. Then, m1, m2, and m3 are input into the interpolation function constructed in step S51 to obtain the predicted short-circuit current value.

[0141] Furthermore, the positive and negative sign combinations defined in step S41 are applied to the normal operation state in step S52. The sign combinations of d1, d2, and d3 determine the category to which the state vector belongs, and the category corresponds to the constraint triggering order. The order in which the iterative evolution crosses the current-limiting boundary in step S4 is the constraint triggering order. In active distribution network short-circuit monitoring, the first triggered constraint is identified from the triggering order, and the current-limiting boundary corresponding to the first triggered constraint is the dominant current-limiting boundary. When the dominant current-limiting boundary is the first current-limiting boundary, it corresponds to the control constraint dominant short-circuit current; when the dominant current-limiting boundary is the second current-limiting boundary, it corresponds to the energy constraint dominant short-circuit current; and when the dominant current-limiting boundary is the third current-limiting boundary, it corresponds to the thermal constraint dominant short-circuit current.

[0142] It should be noted that the application of short-circuit current prediction and dominant current-limiting boundary in active distribution network short-circuit monitoring is to assess the inverter's short-circuit capability and formulate protection strategies. The short-circuit current prediction is used to configure the operating values ​​of relay protection devices; if the predicted value exceeds the rated capacity of the protection device, the protection settings need to be adjusted or the protection device replaced. The dominant current-limiting boundary is used to identify weaknesses in the inverter's short-circuit capability. When the dominant current-limiting boundary is a control constraint boundary, the short-circuit capability is improved by adjusting the controller parameters. When the dominant current-limiting boundary is an energy constraint boundary, the short-circuit capability is improved by increasing the DC capacitor value. When the dominant current-limiting boundary is a thermal constraint boundary, the short-circuit capability is improved by improving heat dissipation conditions or reducing the operating temperature.

[0143] Example 2 is an embodiment of the present invention, which provides an active distribution network short-circuit parameter online monitoring system, comprising:

[0144] The excitation control module is used to control the excitation disturbance applied to the distributed power source;

[0145] The data acquisition module is used to collect operational data under excitation and disturbance conditions.

[0146] The phase space reconstruction module is used to construct state vectors by delaying the embedding of running data, and to record the sequence of state vectors at different times as evolutionary trajectories;

[0147] The current limiting trigger point identification module is used to identify current limiting trigger points from the evolution trajectory;

[0148] The boundary modeling module is used to extract extreme points from the current limiting trigger points of multiple operating conditions and fit the extreme points to obtain the current limiting boundary.

[0149] The trajectory classification module is used to calculate the directed distance from the evolution trajectory to the flow-limiting boundary, classify the evolution trajectory according to the change of the directed distance, and establish a state evolution model for each type of trajectory;

[0150] The short-circuit simulation module is used to apply short-circuit disturbances to the test condition to obtain the initial state, select the state evolution model according to the category of the initial state, use the model to iteratively evolve to the steady state, and extract the short-circuit current characteristics from the evolution process.

[0151] The mapping construction module is used to calculate the normalized directed distance from the initial state to the current-limiting boundary as a margin, and to establish a mapping relationship between the margin and the short-circuit current characteristics.

[0152] The online monitoring module is used to output the predicted short-circuit current and the dominant current-limiting boundary.

[0153] This embodiment also provides an electronic device applicable to an online monitoring method for short-circuit parameters in an active distribution network, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the online monitoring method for short-circuit parameters in an active distribution network as proposed in the above embodiment.

[0154] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements an online monitoring method for active distribution network short-circuit parameters as proposed in the above embodiments.

[0155] The storage medium proposed in this embodiment belongs to the same inventive concept as the method for online monitoring of short-circuit parameters in an active distribution network proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0156] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0157] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for online monitoring of short-circuit parameters in an active distribution network, characterized in that, include: An excitation disturbance is applied to the distributed power source, and the operating data under the excitation disturbance is collected. The operating data is then embedded with a delayed state vector, and the sequence of the state vectors at different times is recorded as an evolution trajectory. The current limiting trigger point is identified from the evolution trajectory. The current limiting trigger point is repeatedly identified under multiple operating conditions. Extreme points are extracted from the current limiting trigger points under multiple operating conditions, and the current limiting boundary is obtained by fitting the extreme points. Calculate the directed distance from the evolution trajectory to the flow-limiting boundary, classify the evolution trajectory according to the change of the directed distance, and establish a state evolution model for each type of trajectory; A short-circuit disturbance is applied to the test condition to obtain an initial state. The state evolution model is selected according to the category to which the initial state belongs. The model is used to iteratively evolve to a steady state, and the short-circuit current characteristics are extracted from the evolution process. Calculate the normalized directed distance from the initial state to the current limiting boundary as a margin, establish the mapping relationship between the margin and the short-circuit current characteristics, and output the predicted short-circuit current value and the dominant current limiting boundary. The time series of parameters in the multiple physical domains include a first electrical parameter time series, a second electrical parameter time series, and a thermal parameter time series; A first embedding vector is obtained by delaying the embedding of the first electrical parameter time series, a second embedding vector is obtained by delaying the embedding of the second electrical parameter time series, and a third embedding vector is obtained by delaying the embedding of the thermal parameter time series. The first embedding vector, the second embedding vector, and the third embedding vector are concatenated to construct an extended phase space state vector. Extract from the current limiting trigger point the first extreme point set, the second extreme point set, and the third extreme point set in the extended phase space where the first embedded vector dimension reaches an extreme value; By fitting the first set of extreme points, the second set of extreme points, and the third set of extreme points respectively, a first current-limiting boundary, a second current-limiting boundary, and a third current-limiting boundary are obtained in the extended phase space. The first current-limiting boundary, the second current-limiting boundary, and the third current-limiting boundary together form a constrained feasible region.

2. The method for online monitoring of short-circuit parameters in an active distribution network as described in claim 1, characterized in that, The operational data includes time series of parameters from multiple physical domains; The method of constructing a state vector by delaying the embedding of the running data includes: delaying the embedding of each parameter time series in the parameter time series of the multiple physical domains to obtain an embedding vector, and concatenating the embedding vectors to construct the state vector; The step of extracting extreme points from the current limiting trigger points of multiple operating conditions includes: extracting a subset of extreme points where each dimension of the embedding vector reaches an extreme value from the current limiting trigger points; The process of fitting the extreme points to obtain the current limiting boundary includes: fitting the extreme point subsets to obtain the current limiting boundary, thereby obtaining multiple current limiting boundaries.

3. The method for online monitoring of short-circuit parameters in an active distribution network as described in claim 2, characterized in that, When the time series of the first electrical parameter is delayed and embedded, the first characteristic time scale is used as the delay time; when the time series of the second electrical parameter is delayed and embedded, the second characteristic time scale is used as the delay time; and when the time series of the thermal parameter is delayed and embedded, the third characteristic time scale is used as the delay time. The first characteristic time scale is determined based on the physical response characteristics of the first electrical parameter, the second characteristic time scale is determined based on the physical response characteristics of the second electrical parameter, and the third characteristic time scale is determined based on the physical response characteristics of the thermal parameter.

4. The method for online monitoring of short-circuit parameters in an active distribution network as described in claim 3, characterized in that, Calculate the first directed distance from each trajectory point on the evolution trajectory to the first flow-limiting boundary, the second directed distance to the second flow-limiting boundary, and the third directed distance to the third flow-limiting boundary; Identify the first sign change moment when the first directed distance changes from positive to negative, the second sign change moment when the second directed distance changes from positive to negative, and the third sign change moment when the third directed distance changes from positive to negative. The first sign change moment is the moment when the evolution trajectory crosses the first current-limiting boundary, the second sign change moment is the moment when the evolution trajectory crosses the second current-limiting boundary, and the third sign change moment is the moment when the evolution trajectory crosses the third current-limiting boundary. The triggering order of the current limiting boundary is determined based on the chronological order of the first symbol change, the second symbol change, and the third symbol change, and evolution trajectories with the same triggering order are grouped into the same category.

5. The method for online monitoring of short-circuit parameters in an active distribution network as described in claim 4, characterized in that, Calculate the directed distance from the initial state to the flow-limiting boundary, determine the category to which the initial state belongs based on the combination of positive and negative signs of the directed distance, and select the state evolution model for that category; The state evolution model is used to perform a single-step evolution of the initial state to obtain the state at the next moment; Calculate the directed distance from the next time-instance state to the current-limiting boundary; When the directional distance sign changes, it is determined that the flow-limiting boundary has been crossed; The category is redefined based on the combination of directed distance symbols after the crossing; Switch to a new category of state evolution model; The state at the next moment is repeatedly subjected to single-step evolution, directed distance calculation, traversal determination, category determination, and model switching for at least one round until the difference between two adjacent evolution states is less than a preset convergence threshold.

6. The method for online monitoring of short-circuit parameters in an active distribution network as described in claim 5, characterized in that, The process of establishing the mapping relationship between the margin and the short-circuit current characteristics includes: constructing an interpolation function by performing multidimensional interpolation on the margin and the short-circuit current characteristics for multiple operating conditions; The output short-circuit current prediction value and dominant current limiting boundary include: collecting operating data during normal operation of the distributed power source; constructing a state vector by delaying the embedding of the operating data; calculating the normalized directed distance from the state vector to the current limiting boundary as a margin; inputting the margin into the interpolation function to obtain the short-circuit current prediction value; calculating the directed distance from the state vector to the current limiting boundary; determining the category to which the state vector belongs based on the combination of positive and negative signs of the directed distance; identifying the first triggered current limiting boundary from the triggering order of the category as the dominant current limiting boundary; and outputting the short-circuit current prediction value and the dominant current limiting boundary.

7. An active distribution network short-circuit parameter online monitoring system, employing the active distribution network short-circuit parameter online monitoring method as described in any one of claims 1 to 6, characterized in that, include: The excitation control module is used to control the excitation disturbance applied to the distributed power source; The data acquisition module is used to collect operational data under excitation and disturbance conditions. The phase space reconstruction module is used to construct state vectors by delaying the embedding of the running data, and record the sequence of state vectors at different times as evolutionary trajectories; A current limiting trigger point identification module is used to identify current limiting trigger points from the evolution trajectory; The boundary modeling module is used to extract extreme points from the current limiting trigger points of multiple operating conditions and fit the extreme points to obtain the current limiting boundary; The trajectory classification module is used to calculate the directed distance from the evolution trajectory to the flow-limiting boundary, classify the evolution trajectory according to the change of the directed distance, and establish a state evolution model for each type of trajectory; The short-circuit simulation module is used to apply a short-circuit disturbance to the test condition to obtain an initial state, select the state evolution model according to the category of the initial state, use the model to iteratively evolve to a steady state, and extract short-circuit current characteristics from the evolution process. The mapping construction module is used to calculate the normalized directed distance from the initial state to the current limiting boundary as a margin, and to establish a mapping relationship between the margin and the short-circuit current characteristics. The online monitoring module is used to output the predicted short-circuit current and the dominant current-limiting boundary.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the online monitoring method for short-circuit parameters of an active distribution network as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the online monitoring method for short-circuit parameters of an active distribution network as described in any one of claims 1 to 6.

Images