Active power distribution network equivalent modeling method and system based on multi-target parameter identification

Abstract

The application discloses a method and system for active power distribution network equivalent modeling based on multi-target parameter identification, takes the low-voltage ride-through time of distributed photovoltaic as a clustering characteristic index, adopts a hierarchical clustering algorithm to complete distributed photovoltaic clustering, and determines an optimal clustering number through a contour coefficient; extracts to-be-determined parameters from full-amount electrical parameters to form a to-be-screened parameter set, calculates the trajectory sensitivity of each parameter based on a power distribution network detailed model, forms a to-be-identified parameter set from parameters meeting the conditions, and directly aggregates the remaining parameters through a clustering result; takes active power and reactive power errors at a grid-connected point as optimization targets, adopts a multi-target improved marine predation algorithm to optimize and identify the to-be-identified parameters, and forms an active power distribution network equivalent model according to the clustering result and the identification result. In the equivalent modeling process, the application introduces a multi-target parameter identification and key parameter screening mechanism, can improve the fitting precision of the equivalent model on the steady-state and transient-state characteristics of the active power distribution network, reduces the parameter identification complexity, and improves the modeling efficiency and the calculation convergence speed.

Description

Technical Field

[0001] This invention belongs to the field of active distribution network model parameter identification and equivalent modeling technology in power systems, and particularly relates to an active distribution network equivalent modeling method and system based on multi-objective parameter identification. Background Technology

[0002] With the integration of a high proportion of inverter-based distributed generation (IBDG, such as photovoltaic and wind power) into the distribution network, the traditional distribution network is gradually evolving into an active distribution network (ADN) with various power electronic devices interacting with loads. In large-scale power grid transient stability analysis and simulation, to reduce computational scale, the active distribution network is typically simplified to a Thevenin equivalent circuit parallel dynamic equivalent model (such as a ZIP parallel induction motor model or an IBDG aggregation equivalent machine). Dynamic equivalent modeling of active distribution networks mainly includes two key steps: first, determining the equivalent model structure by grouping and aggregating parameters based on the dynamic characteristics of distributed generation and loads; and second, identifying the key electrical parameters of the equivalent model.

[0003] Several dynamic equivalent methods for active distribution networks already exist in the current technology. For example, existing technology 1 ("Dynamic Equivalents of Active Distribution Networks Considering IBDG Transient Characteristics", IEEE, Shanhua Hu et al., 2024) proposes to introduce the low-voltage ride-through (LVRT) exit time of IBDGs as an indicator to characterize their transient characteristics. It uses the AGNES (Agglomerative Nesting) hierarchical clustering algorithm to cluster IBDGs in the distribution network, determines the optimal number of clusters using multiple evaluation indicators, and then aggregates the parameters of IBDGs within the same cluster to obtain the ADN dynamic equivalent model. This method considers the impact of IBDG transient control characteristics on the accuracy of the equivalent model, which is an improvement over the traditional aggregation method that only weights by geographical location or capacity.

[0004] Another existing technology 2 (Dynamic equivalent modelling for active distributed network considering adjustable loads charging characteristics”, IETGeneration, Transmission & Distribution, Jingwen Wang et al., 2024) uses a ZIP load parallel induction motor as the equivalent structure of ADN and uses an improved deep dual-Q network (MDDQN) to perform single-object identification of the equivalent model parameters, while realizing the internal subdivision identification of adjustable loads such as electric vehicles.

[0005] In addition, trajectory sensitivity analysis for screening key / non-key parameters and the use of metaheuristic intelligent algorithms (such as particle swarm optimization, genetic algorithm, ocean predator algorithm, etc.) for power system parameter identification are also well-known parameter processing methods in this field.

[0006] However, the aforementioned prior art has at least the following defects and shortcomings: The clustering and aggregation dimension does not incorporate parameter sensitivity analysis: Although the existing technology 1 performs IBDG clustering and parameter aggregation based on LVRT exit time, it does not introduce trajectory sensitivity to critically screen the parameters of the equivalent model, resulting in an excessively high dimension of the parameters to be estimated during subsequent parameter identification, which is easily affected by noise from non-critical parameters, reducing identification efficiency and stability. Parameter identification is a single-objective optimization and is prone to getting trapped in local optima: Existing techniques 1 and traditional methods mostly use least squares or single-objective heuristic algorithms for parameter fitting, aiming only at minimizing the single variance of active or reactive power at the grid connection point, making it difficult to consider the accuracy of dynamic response on both the P and Q sides, resulting in large equivalent errors under large disturbances; Existing technique 2, although introducing deep reinforcement learning for identification, is still a single-objective reward function optimization, and DRL training relies on a large number of samples, limiting its generalization ability. At the same time, conventional metaheuristic algorithms (such as standard PSO, GA, and basic MPA) have problems such as slow convergence speed and easy premature convergence and getting trapped in local optima, resulting in insufficient accuracy when directly used for high-dimensional nonlinear equivalent parameter identification; Lack of multi-objective Pareto compromise decision-making mechanism: Most existing active distribution network equivalent parameter identification methods directly output a single set of optimal solutions, without considering the conflict between P and Q objectives to generate the Pareto front and select the compromise solution that best balances the accuracy of both sides based on fuzzy comprehensive evaluation. This results in poor adaptability of the equivalent model under different operating conditions and disturbance levels.

[0007] Therefore, there is an urgent need for a dynamic equivalent modeling method for active distribution networks that can comprehensively consider IBDG transient characteristic grouping, parameter sensitivity dimensionality reduction screening, and multi-objective robust parameter identification and fuzzy decision-making, so as to improve the accuracy and versatility of the equivalent model of active distribution networks under large disturbances. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides an equivalent modeling method and system for active distribution networks based on multi-objective parameter identification. Using the low-voltage ride-through time of distributed photovoltaic (PV) systems as a clustering characteristic indicator, a hierarchical clustering algorithm is employed to cluster the PV systems, and the optimal number of clusters is determined using the silhouette coefficient. Undetermined parameters are extracted from the total electrical parameters to form a set of parameters to be screened. The trajectory sensitivity of each parameter is calculated based on a detailed distribution network model. Parameters meeting the criteria form a set of parameters to be identified, while the remaining parameters are directly aggregated using the clustering results. Using the active and reactive power errors at the grid connection point as optimization objectives, a multi-objective improved marine predation algorithm is used to optimize and identify the parameters to be identified. An equivalent model of the active distribution network is formed based on the clustering and identification results. This invention introduces a multi-objective parameter identification and key parameter screening mechanism in the equivalent modeling process, which can improve the fitting accuracy of the equivalent model to the steady-state and transient characteristics of the active distribution network, reduce the complexity of parameter identification, and improve modeling efficiency and computational convergence speed.

[0009] The present invention adopts the following technical solution.

[0010] An equivalent modeling method for active distribution networks based on multi-objective parameter identification includes: Based on the active distribution network, the equivalent boundary is determined, all electrical parameters are collected, and a detailed model of the distribution network including distributed photovoltaic power is constructed. Power flow and transient calculations are performed on a detailed model of an active distribution network, and power and voltage values ​​during steady-state and transient periods are collected at equivalent boundaries. The distributed photovoltaic system is grouped using the low-voltage ride-through time of each distributed photovoltaic system as a clustering characteristic index, and the optimal number of clusters is determined by the silhouette coefficient. Undetermined parameters are extracted from the full set of electrical parameters to form a set of parameters to be screened; based on the detailed model of the distribution network, the trajectory sensitivity of each parameter in the set of parameters to be screened is calculated, and the parameters are screened according to the trajectory sensitivity. The parameters that meet the conditions form a set of parameters to be identified, and the remaining parameters are directly aggregated through the clustering results. Based on steady-state and transient time-series data at the equivalent boundary, a parameter identification objective function is constructed. Based on the parameter identification objective function, a multi-objective improved marine predation algorithm is used to identify the parameters to be identified in the parameter set. Based on the clustering results and parameter identification results, an active power distribution network multi-machine equivalent model is constructed.

[0011] More preferably, the distributed photovoltaic (PV) systems are grouped using the low-voltage ride-through time of each PV system as a clustering characteristic index, and the optimal number of clusters is determined by the silhouette coefficient, including: The moment when the terminal voltage of each distributed photovoltaic system recovers to a preset voltage threshold during low-voltage ride-through is obtained, and this moment is used as the exit time of low-voltage ride-through to construct a cluster feature vector; A hierarchical clustering algorithm is used to calculate the Euclidean distance between samples and the distance between clusters based on the cluster feature vectors. The clustering tree is generated by iteratively merging the clusters with the smallest distance. Calculate the overall profile coefficient corresponding to different numbers of clusters in the clustering tree, determine the number of clusters corresponding to the largest overall profile coefficient as the optimal number of clusters, and divide each distributed photovoltaic power generation into multiple clusters based on the optimal number of clusters.

[0012] More preferably, the step of calculating the Euclidean distance between samples and the inter-cluster distance based on the cluster feature vector includes: The absolute value of the difference between the exit low-voltage ride-through times of any two distributed photovoltaic systems is taken as the Euclidean distance between the samples. The inter-cluster distance is calculated by dividing the product of the number of samples in any two clusters by the sum of the number of samples in those two clusters, and then multiplying by the square of the norm of the difference between the means of samples in those two clusters. The calculation of the overall profile coefficient corresponding to different numbers of clusters in the clustering tree includes: For each sample in the clustering tree, the average distance from the sample to other samples in its own cluster is calculated as the first distance, and the average distance from the sample to all samples in the nearest other cluster is calculated as the second distance; The difference between the second distance and the first distance is divided by the maximum value of the first distance and the second distance to obtain the contour coefficient of the sample; The average of the silhouette coefficients of all samples is taken as the overall silhouette coefficient.

[0013] More preferably, the trajectory sensitivity is calculated for each parameter in the parameter set to be screened, and the parameters are screened based on the trajectory sensitivity. Parameters that meet the conditions constitute the parameter set to be identified, and the remaining parameters are directly aggregated through the clustering results, including: Calculate the average trajectory sensitivity of each parameter in the parameter set to be screened to multiple electrical quantities of the active distribution network, and sum the average trajectory sensitivity of each electrical quantity; The parameters whose sum of average trajectory sensitivities is greater than a preset sensitivity threshold are determined as undetermined parameters, forming the set of parameters to be identified; For the remaining parameters, based on the clustering results, the parameters of distributed photovoltaics within the same cluster are aggregated using the rated capacity weighted average method to obtain the aggregated parameters of the equivalent unit.

[0014] More preferably, the average trajectory sensitivity of each parameter in the parameter set to be screened to multiple electrical quantities of the active distribution network is calculated, and the average trajectory sensitivity of each electrical quantity is summed, including: For any parameter in the set of parameters to be screened, calculate the instantaneous trajectory sensitivity of the parameter at each sampling time within a preset observation period. The instantaneous trajectory sensitivity is the ratio of the relative change rate of each electrical quantity to the relative change rate of the parameter when the parameter changes. Integrate the instantaneous trajectory sensitivity within the preset observation period and calculate the average value to obtain the average trajectory sensitivity of the parameter for each electrical quantity; The method of aggregating parameters of distributed photovoltaic systems within the same cluster using the rated capacity weighted average method to obtain aggregated parameters for equivalent units includes: The aggregate parameter of the equivalent unit is obtained by summing the products of the rated capacity of each distributed photovoltaic unit within the same cluster and the corresponding parameter value, and then dividing by the sum of the rated capacities of each distributed photovoltaic unit within the same cluster.

[0015] More preferably, the construction of the parameter identification objective function based on steady-state and transient time-series data at the equivalent boundary includes: The parameter identification objective function is constructed by taking the root mean square error of active power and the root mean square error of reactive power at the grid connection point of the active distribution network as two independent objective terms. The root mean square error of active power is the root mean square of the sum of squares of the differences between the reference value of active power at the equivalent boundary and the calculated value of the equivalent model. The root mean square error of reactive power is the root mean square of the sum of squares of the differences between the reactive power reference value and the calculated value of the equivalent model at the equivalent boundary.

[0016] More preferably, a multi-objective improved marine predation algorithm is used to identify the parameters to be identified in the parameter set, including: An initial population uniformly covering the solution space is generated using the optimal point set theory. A fast non-dominated sort is performed on the initial population. The initial population is divided into multiple non-dominated layers according to the dominance relationship. Individuals are selected from the first non-dominated layer to construct an elite matrix. An adaptive step size control factor is introduced, and the optimization process is divided into multiple stages according to the current iteration number. Different motion strategies are used in different stages to update the position of individuals in the population until the maximum number of iterations is reached, and the final Pareto optimal parameter set is output.

[0017] More preferably, the introduction of an adaptive step size control factor divides the optimization process into multiple stages based on the current iteration number, and uses different movement strategies to update the individual positions of the population at different stages, including: The adaptive step size control factor is calculated based on the current iteration number, the maximum iteration number, and the preset initial and final step size values. When the current iteration count is less than one-third of the maximum iteration count, the prey position is updated using a Brownian motion random vector for global exploration. When the current iteration number is greater than or equal to one-third of the maximum iteration number and less than two-thirds of the maximum iteration number, the first half of the population is explored using a Levy flight random vector, and the second half of the population is developed using a Brownian motion random vector. When the current iteration number is greater than or equal to two-thirds of the maximum iteration number, the Levy flight random vector is used to develop all populations to approach the global optimum region.

[0018] More preferably, after identifying the parameters to be identified in the parameter set using the multi-objective improved marine predation algorithm, the method further includes: Obtain the maximum and minimum root mean square error of each electrical quantity in the Pareto optimal parameter set of the parameter identification output; For any set of candidate parameters in the Pareto optimal parameter set, the difference between the maximum value of the root mean square error of the electrical quantity and the function value of the candidate parameter on the root mean square error of the electrical quantity is divided by the difference between the maximum value and the minimum value to obtain the linear membership degree of the candidate parameter on the root mean square error of the electrical quantity. The average of the linear membership degree of the candidate parameter on the root mean square error of active power and the linear membership degree on the root mean square error of reactive power is used to obtain the overall satisfaction of the candidate parameter. The candidate parameter with the highest overall satisfaction is selected as the optimal compromise parameter. The optimal compromise parameter and the remaining parameters after parameter aggregation through the clustering results are used to generate the active distribution network multi-machine equivalent model.

[0019] This invention also proposes an equivalent modeling system for active distribution networks based on multi-objective parameter identification, comprising a model building module, a data acquisition module, a clustering module, a parameter filtering and aggregation module, a parameter identification module, and an equivalent modeling module, characterized in that: The model building module determines the equivalent boundary based on the active distribution network, collects all electrical parameters, and constructs a detailed model of the distribution network including distributed photovoltaics. The data acquisition module performs power flow and transient calculations on the detailed model of the active distribution network, and collects the power and voltage values ​​during steady-state and transient periods at the equivalent boundary. The clustering module uses the low-voltage ride-through time of each distributed photovoltaic (PV) unit as a clustering characteristic index to group distributed PV units, and determines the optimal number of clusters through the silhouette coefficient. The parameter filtering and aggregation module extracts undetermined parameters from the full set of electrical parameters to form a set of parameters to be filtered; based on the detailed model of the distribution network, it calculates the trajectory sensitivity of each parameter in the set of parameters to be filtered, filters the parameters according to the trajectory sensitivity, and forms a set of parameters to be identified by the parameters that meet the conditions; the remaining parameters are directly aggregated through the clustering results. The parameter identification module constructs a parameter identification objective function based on steady-state and transient time-series data at the iso-boundary. Based on the parameter identification objective function, a multi-objective improved marine predation algorithm is used to identify the parameters to be identified in the parameter set. The equivalent modeling module constructs an equivalent model of multiple machines in the active distribution network based on the clustering results and parameter identification results.

[0020] The present invention also proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method steps described above.

[0021] The present invention also proposes a terminal, including a processor and a storage medium: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps according to the method described above.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention uses trajectory sensitivity analysis to quantify the degree of influence of each parameter on the transient state, achieves accurate separation of key and non-key parameters, reduces the computational complexity of parameter identification, and significantly improves the computational convergence speed.

[0023] 2. This invention uses an improved marine predation algorithm to optimize the fitting accuracy of active and reactive forces, solving the problem of inherent accuracy imbalance in single-target parameter identification. Combined with the fuzzy comprehensive evaluation method, the Pareto optimal compromise parameter is scientifically selected, thereby improving the fitting accuracy of the equivalent model. Attached Figure Description

[0024] Figure 1 This is a flowchart of the active power distribution network equivalent modeling method based on multi-objective parameter identification of the present invention; Figure 2 This is a block diagram of the equivalent modeling structure of the active power distribution network based on multi-objective parameter identification of the present invention; Figure 3This is a detailed simulation model of the active power distribution network of the present invention; Figure 4 This is a comparison diagram of the active power waveforms between the detailed simulation model and the equivalent simulation model of the active power distribution network of the present invention; Figure 5 This is a comparison diagram of reactive power waveforms between the detailed simulation model and the equivalent simulation model of the active power distribution network of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0026] like Figure 1 As shown, this invention proposes an equivalent modeling method for active distribution networks based on multi-objective parameter identification, specifically including the following steps: Based on the active distribution network, the equivalent boundary is determined, all electrical parameters are collected, and a detailed model of the distribution network including distributed photovoltaic power is constructed. Power flow and transient calculations are performed on a detailed model of an active distribution network, and power and voltage values ​​during steady-state and transient periods are collected at equivalent boundaries. The distributed photovoltaic system is grouped using the low-voltage ride-through time of each distributed photovoltaic system as a clustering characteristic index, and the optimal number of clusters is determined by the silhouette coefficient. The moment when the terminal voltage of each distributed photovoltaic system recovers to a preset voltage threshold during low-voltage ride-through is obtained, and this moment is used as the exit time of low-voltage ride-through to construct a cluster feature vector; A hierarchical clustering algorithm is used to calculate the Euclidean distance between samples and the distance between clusters based on the cluster feature vectors. The clustering tree is generated by iteratively merging the clusters with the smallest distance. Calculate the overall profile coefficient corresponding to different numbers of clusters in the clustering tree, determine the number of clusters corresponding to the largest overall profile coefficient as the optimal number of clusters, and divide each distributed photovoltaic power generation into multiple clusters based on the optimal number of clusters.

[0027] The calculation of Euclidean distance between samples and inter-cluster distance based on the cluster feature vectors includes: The absolute value of the difference between the exit low-voltage ride-through times of any two distributed photovoltaic systems is taken as the Euclidean distance between the samples. The inter-cluster distance is calculated by dividing the product of the number of samples in any two clusters by the sum of the number of samples in those two clusters, and then multiplying by the square of the norm of the difference between the means of samples in those two clusters. The calculation of the overall profile coefficient corresponding to different numbers of clusters in the clustering tree includes: For each sample in the clustering tree, the average distance from the sample to other samples in its own cluster is calculated as the first distance, and the average distance from the sample to all samples in the nearest other cluster is calculated as the second distance; The difference between the second distance and the first distance is divided by the maximum value of the first distance and the second distance to obtain the contour coefficient of the sample; The average of the silhouette coefficients of all samples is taken as the overall silhouette coefficient.

[0028] Undetermined parameters are extracted from the full set of electrical parameters to form a set of parameters to be screened; based on the detailed model of the distribution network, the trajectory sensitivity of each parameter in the set of parameters to be screened is calculated, and the parameters are screened according to the trajectory sensitivity. The parameters that meet the conditions form a set of parameters to be identified, and the remaining parameters are directly aggregated through the clustering results. Calculate the average trajectory sensitivity of each parameter in the parameter set to be screened to multiple electrical quantities of the active distribution network, and sum the average trajectory sensitivity of each electrical quantity; The parameters whose sum of average trajectory sensitivities is greater than a preset sensitivity threshold are determined as undetermined parameters, forming the set of parameters to be identified; For the remaining parameters, based on the clustering results, the parameters of distributed photovoltaics within the same cluster are aggregated using the rated capacity weighted average method to obtain the aggregated parameters of the equivalent unit.

[0029] Calculate the average trajectory sensitivity of each parameter in the parameter set to be screened to multiple electrical quantities in the active distribution network, and sum the average trajectory sensitivity of each electrical quantity, including: For any parameter in the set of parameters to be screened, calculate the instantaneous trajectory sensitivity of the parameter at each sampling time within a preset observation period. The instantaneous trajectory sensitivity is the ratio of the relative change rate of each electrical quantity to the relative change rate of the parameter when the parameter changes. Integrate the instantaneous trajectory sensitivity within the preset observation period and calculate the average value to obtain the average trajectory sensitivity of the parameter for each electrical quantity; The method of aggregating parameters of distributed photovoltaic systems within the same cluster using the rated capacity weighted average method to obtain aggregated parameters for equivalent units includes: The aggregate parameter of the equivalent unit is obtained by summing the products of the rated capacity of each distributed photovoltaic unit within the same cluster and the corresponding parameter value, and then dividing by the sum of the rated capacities of each distributed photovoltaic unit within the same cluster.

[0030] Based on steady-state and transient time-series data at the equivalent boundary, a parameter identification objective function is constructed. Based on the parameter identification objective function, a multi-objective improved marine predation algorithm is used to identify the parameters to be identified in the parameter set. Based on the clustering results and parameter identification results, an active power distribution network multi-machine equivalent model is constructed.

[0031] The parameter identification objective function is constructed based on the steady-state and transient time-series data at the equivalent boundary, including: The parameter identification objective function is constructed by taking the root mean square error of active power and the root mean square error of reactive power at the grid connection point of the active distribution network as two independent objective terms. The root mean square error of active power is the root mean square of the sum of squares of the differences between the reference value of active power at the equivalent boundary and the calculated value of the equivalent model. The root mean square error of reactive power is the root mean square of the sum of squares of the differences between the reactive power reference value and the calculated value of the equivalent model at the equivalent boundary.

[0032] An improved multi-objective marine predation algorithm is used to identify parameters in the parameter set to be identified, including: An initial population uniformly covering the solution space is generated using the optimal point set theory. A fast non-dominated sort is performed on the initial population. The initial population is divided into multiple non-dominated layers according to the dominance relationship. Individuals are selected from the first non-dominated layer to construct an elite matrix. An adaptive step size control factor is introduced, and the optimization process is divided into multiple stages according to the current iteration number. Different motion strategies are used in different stages to update the position of individuals in the population until the maximum number of iterations is reached, and the final Pareto optimal parameter set is output.

[0033] The introduction of an adaptive step size control factor divides the optimization process into multiple stages based on the current iteration number, employing different movement strategies to update the individual positions in the population at different stages, including: The adaptive step size control factor is calculated based on the current iteration number, the maximum iteration number, and the preset initial and final step size values. When the current iteration count is less than one-third of the maximum iteration count, the prey position is updated using a Brownian motion random vector for global exploration. When the current iteration number is greater than or equal to one-third of the maximum iteration number and less than two-thirds of the maximum iteration number, the first half of the population is explored using a Levy flight random vector, and the second half of the population is developed using a Brownian motion random vector. When the current iteration number is greater than or equal to two-thirds of the maximum iteration number, the Levy flight random vector is used to develop all populations to approach the global optimum region.

[0034] After the multi-objective improved marine predation algorithm identifies the parameters to be identified in the parameter set, the method further includes: Obtain the maximum and minimum root mean square error of each electrical quantity in the Pareto optimal parameter set of the parameter identification output; For any set of candidate parameters in the Pareto optimal parameter set, the difference between the maximum value of the root mean square error of the electrical quantity and the function value of the candidate parameter on the root mean square error of the electrical quantity is divided by the difference between the maximum value and the minimum value to obtain the linear membership degree of the candidate parameter on the root mean square error of the electrical quantity. The average of the linear membership degree of the candidate parameter on the root mean square error of active power and the linear membership degree on the root mean square error of reactive power is used to obtain the overall satisfaction of the candidate parameter. The candidate parameter with the highest overall satisfaction is selected as the optimal compromise parameter. The optimal compromise parameter and the remaining parameters after parameter aggregation through the clustering results are used to generate the active distribution network multi-machine equivalent model.

[0035] This invention also proposes an equivalent modeling system for active distribution networks based on multi-objective parameter identification, comprising a model building module, a data acquisition module, a clustering module, a parameter filtering and aggregation module, a parameter identification module, and an equivalent modeling module, characterized in that: The model building module determines the equivalent boundary based on the active distribution network, collects all electrical parameters, and constructs a detailed model of the distribution network including distributed photovoltaics. The data acquisition module performs power flow and transient calculations on the detailed model of the active distribution network, and collects the power and voltage values ​​during steady-state and transient periods at the equivalent boundary. The clustering module uses the low-voltage ride-through time of each distributed photovoltaic (PV) unit as a clustering characteristic index to group distributed PV units, and determines the optimal number of clusters through the silhouette coefficient. The parameter filtering and aggregation module extracts undetermined parameters from the full set of electrical parameters to form a set of parameters to be filtered; based on the detailed model of the distribution network, it calculates the trajectory sensitivity of each parameter in the set of parameters to be filtered, filters the parameters according to the trajectory sensitivity, and forms a set of parameters to be identified by the parameters that meet the conditions; the remaining parameters are directly aggregated through the clustering results. The parameter identification module constructs a parameter identification objective function based on steady-state and transient time-series data at the iso-boundary. Based on the parameter identification objective function, a multi-objective improved marine predation algorithm is used to identify the parameters to be identified in the parameter set. The equivalent modeling module constructs an equivalent model of multiple machines in the active distribution network based on the clustering results and parameter identification results.

[0036] The present invention also proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method steps described above.

[0037] The present invention also proposes a terminal, including a processor and a storage medium: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps according to the method described above.

[0038] Example 1 like Figure 2 As shown, this embodiment proposes an equivalent modeling method for active distribution networks based on multi-objective parameter identification, including the following steps: Step 1: Using the connection bus between the active distribution network and the upstream main grid as the equivalent boundary node, select the 35kV bus as the grid connection point PCC node. The active and reactive power control strategies for photovoltaic units during low-voltage ride-through (LVRT) in the detailed model of the active distribution network are as follows: During normal operation, distributed photovoltaic systems employ constant active power and constant power factor control. In this embodiment, the power factor is set to 1, therefore the reactive current reference value is 0.

[0039] Active power control strategy: (2) in This refers to the active current during LVRT. The active current control coefficient is 1; The active current control coefficient is 2; This refers to the active current before the fault. This refers to the amplitude of the terminal voltage. This is the active current setting value.

[0040] Reactive power control strategy: (3) In the formula, where This refers to the reactive current during LVRT. The reactive current control coefficient is 1; The reactive current control coefficient is 2; This refers to the reactive current prior to the fault. This refers to the amplitude of the terminal voltage. To reach the low-voltage ride-through threshold, a value of 0.9 pu is used. This is the reactive current setting value.

[0041] In the above formula , , , Together they constitute the photovoltaic low-voltage ride-through control parameter set Furthermore, the equivalent impedance parameters, induction motor parameters, ZIP static load parameters, and reactive power compensation parameters are all included in the parameter set to be screened, namely: (4) in, For the network equivalent impedance parameter set, This represents the equivalent impedance of an active power distribution network. This represents the equivalent impedance of an active power distribution network. Here is the parameter set for the induction motor, where, This represents the stator resistance of the induction motor; Indicates rotor resistance; Indicates stator reactance; Indicates rotor reactance; Indicates the magnetizing reactance; Represents the rotor's inertia time constant; This indicates the initial slip of the induction motor. This is the ZIP static load parameter set, where, This indicates the proportion of active constant impedance loads. Indicates the proportion of active constant current load; This indicates the proportion of reactive constant impedance loads. This indicates the proportion of reactive constant current load. For reactive power compensation parameter set, This represents the equivalent reactive power compensation capacity.

[0042] Step Two: Perform power flow calculations on the detailed model of the active distribution network using the Newton-Raphson method. Transient calculations are performed on the detailed model of the active distribution network. A three-phase fault is introduced at the grid connection point to induce a low-voltage ride-through state. Active power, reactive power, and voltage values ​​at the photovoltaic generator terminals are collected. The detailed model of the active distribution network is shown below. Figure 3 .

[0043] Step 3: The low-voltage ride-through time is defined as when the terminal voltage of the distributed photovoltaic system reaches 0.9 pu. The AGNES hierarchical clustering algorithm is used to group the distributed photovoltaic systems, and the optimal number of clusters is determined by the silhouette coefficient. The specific steps are as follows: Record the moment when the voltage recovers to 0.9 pu after the unit fault is cleared, and iterate through all distributed units to form a cluster feature vector: (5) Where N is the total number of photovoltaic units. Let k be the time it takes for the kth photovoltaic unit to exit the low-voltage ride-through phase.

[0044] Euclidean distance between samples: (6) in, This represents the Euclidean distance between two samples, i and j. These are the low-voltage ride-through times for the i-th and j-th photovoltaic units, respectively.

[0045] Inter-cluster distance: (7) in, The inter-cluster distance between clusters p and q. There are two clusters; This represents the number of samples within the corresponding cluster. This represents the mean of the samples within the corresponding cluster.

[0046] By iteratively searching for the two clusters with the smallest distance in the current cluster set and merging them, the cluster set and the distance between clusters are updated synchronously until all samples are aggregated into a complete cluster, generating a clustering tree.

[0047] Calculate the silhouette coefficient for each segmented sample. Calculate its profile coefficient: (8) in, For the sample The average distance to other samples in its cluster is expressed as: (9) in, It is a cluster The number of samples in; For the sample and The Euclidean distance between them.

[0048] For the sample The average distance to the nearest other cluster sample is expressed as: (10) In the formula, Within the cluster The number of samples in; Cluster Is it except Other clusters besides the one it belongs to.

[0049] For candidate cluster numbers K=2, 3, ... The clustering tree is segmented separately, and the overall profile coefficient is calculated: (11) In the formula The overall profile coefficient when the number of clusters is K; Let S be the total number of distributed photovoltaic (PV) units; S(i) is the silhouette coefficient of the sample corresponding to the i-th distributed PV unit. For the number of candidate clusters... For each segmented cluster tree, calculate the overall profile coefficient for each cluster number. and will The maximum time corresponding The optimal number of clusters was determined.

[0050] Step 4: From the full electrical parameters of the active distribution network detailed model established in Step 1, extract the undetermined parameters involved in the equivalent model (including but not limited to photovoltaic control parameters, induction motor parameters, ZIP load factor, etc.) to form a parameter set to be screened; then, calculate the trajectory sensitivity for each parameter in this parameter set, and screen out the parameters that have the greatest impact on transient response as the parameter set to be identified. For the remaining parameters, directly aggregate the parameters through the clustering results. The detailed steps are as follows: For the j-th parameter in the parameter set to be filtered Define the electrical quantity output by the detailed model of the active distribution network at time t. Instantaneous trajectory sensitivity for: (12) To visually compare the influence of each parameter, the average trajectory sensitivity is calculated, and the expression is: (13) in, This represents the average trajectory sensitivity, i.e., the degree of influence of the j-th model parameter on the i-th output variable; t represents the instantaneous trajectory sensitivity; t is the time of the fault occurrence. To observe the total duration of the transient response.

[0051] Parameters whose sum of the average active power trajectory sensitivity and the average reactive power trajectory sensitivity is greater than a threshold are identified as key parameters that significantly affect the transient characteristics of the distribution network, forming a parameter set to be identified and included in the parameter set in step five. For non-key parameters with low sensitivity, aggregation is performed directly based on the clustering results from step three using a weighted average method according to rated capacity. The calculation formula is as follows: (14) in, The aggregation parameter for the iso-unit; The rated capacity of the i-th device or load; Its corresponding parameters.

[0052] Step 5: Employ a multi-objective improved marine predation algorithm to identify the parameters determined in Step 4. The essence of parameter identification is to find a set of parameters. This ensures that the equivalent model most closely approximates the dynamic response of the detailed distribution network under the same disturbance. The specific steps are as follows: By treating the root mean square error of active power and reactive power at the grid connection point as two independent sub-objectives, a multi-objective optimization problem is constructed: (15) in The vector of parameters to be identified is the set of parameters to be identified determined in step four. The root mean square error of active power. The root mean square error of reactive power is calculated using the following formula: (16) Where N is the number of simulation sampling points; The active and reactive power reference values ​​for the i-th sampling point of the detailed model collected in step two; Parameters to be identified Below, the calculated active and reactive power values ​​at the i-th sampling point of the equivalent model.

[0053] To obtain the optimal solution set, this invention combines an improved marine predation algorithm with fast non-dominated sorting and a reference point-based environment selection strategy. The specific steps are as follows: The initial population covering the uniform solution space is generated using the optimal point set theory, replacing the traditional random initialization, thus improving the algorithm's optimization efficiency. The formula for generating the initial population is as follows: (17) in, Let j be the value of the j-th dimension of the parameter to be identified for the Kth individual in the population. , These are the upper and lower bounds of the j-th dimension parameter, respectively; Indicates taking the decimal part; The optimal point is defined as p, where p is the smallest prime number greater than 2D, D is the dimension of the optimization variable, and k is the population individual index.

[0054] The initial population is constructed as an n×D dimensional prey matrix, expressed as: (18) Perform a fast non-dominated sort on the initial population, dividing the population into multiple non-dominated layers according to dominance relationships. ,in For the Pareto optimal frontier, starting from the first non-dominated layer A single individual is randomly selected as the apex predator, and it is replicated n times to construct an elite matrix, guiding the population to iteratively seek the best. (19) Introduce an improved adaptive step size control factor CF, which enables the algorithm to maintain a large step size in the early stage to facilitate global exploration, and rapidly attenuate the step size in the later stage to accelerate local fine exploitation. The expression is as follows: (20) where, and are respectively the initial value and the final value of CF; r is a uniform random number in [0, 1].

[0055] The position of each individual in the population is updated in three stages according to the current iteration number t.

[0056] Stage 1 (t<T / 3): High-speed ratio global exploration stage, where the prey performs Brownian motion and the predators remain stationary to enhance global exploration. The position of the i-th prey is updated as: (21) Stage 2 (T / 3≤t<2T / 3): The first half of the population uses Lévy flight for exploration, and the second half uses Brownian motion for exploitation. For the first 1 / 2 of the population: (22) For the last 1 / 2 of the population; (23) Stage 3 (t≥2T / 3): All prey use Lévy flight for precise exploitation to quickly approach the global optimal region: (24) where t is the current iteration number, is the current individual vector position, is its corresponding top predator position, is the Brownian motion random vector, is the Lévy flight random vector, is the uniform random vector in [0,1], is the individual vector of the i-th row in the prey matrix; CF is the adaptive step size control factor.

[0057] The iteration is repeated until the maximum number of iterations is reached, and the final non-dominated solution set is output as the identified Pareto optimal parameter set. To determine the optimal compromise parameters, the fuzzy comprehensive evaluation method is adopted for decision-making. For the i-th Pareto solution, the k-th objective function f k has a satisfaction function defined as a linear membership degree: (25) In the formula, are respectively the maximum value and the minimum value of the k-th objective in the Pareto solution set, The objective function value of the i-th Pareto solution on the k-th objective. Let be the satisfaction level of the i-th solution on the k-th objective, with a value range of [0,1].

[0058] Assuming that active power error and reactive power error are equally important, the overall satisfaction level of the i-th solution is: (25) Let be the satisfaction level of the i-th solution with respect to the active power error objective. Let i be the satisfaction level of the i-th solution with respect to the reactive power error target. The overall satisfaction level of the i-th solution also ranges from [0,1]. Select the overall satisfaction level... The largest solution is used as the final identification parameter.

[0059] Figure 4 A comparison chart of active power waveforms during voltage dips at the grid connection point's output voltage setting. Figure 5 The diagram shows a comparison of reactive power fluctuation waveforms at the grid connection point. Simulation results indicate that the established equivalent model and the detailed model exhibit essentially the same transient response to the external power grid during low-voltage crossing, demonstrating good consistency.

[0060] Example 2 This embodiment provides an equivalent modeling system for active power distribution networks based on multi-objective parameter identification. As the hardware platform for implementing the aforementioned equivalent modeling method, this system employs a modular architecture design to decouple complex parameter identification and control logic into independent data processing units, ensuring the system's stability and maintainability in scenarios involving the fusion of multi-source heterogeneous data. Specifically, the system includes: a model building module, a data acquisition module, a clustering module, a parameter filtering and aggregation module, a parameter identification module, and an equivalent modeling module.

[0061] The model building module is used to determine the equivalent boundary based on the active distribution network, collect all electrical parameters, and build a detailed model of the distribution network including distributed photovoltaics.

[0062] Specifically, the model building module, serving as the system's data foundation, is configured to communicate with the distribution network monitoring system or historical database to obtain the topology, equipment nameplate parameters, and control parameters of the area to be equated. Internally, this module includes a boundary resolution unit and a parameter input unit. The boundary resolution unit is responsible for determining the equating boundary nodes based on scheduling requirements, separating the external power grid from the distribution network area to be equated. The parameter input unit inputs all electrical parameters of each photovoltaic unit within the area, including control parameters (such as LVRT control coefficients, active / reactive current command coefficients), line impedance, transformer parameters, induction motor parameters, and ZIP load factors. This data flow preprocessing mechanism provides a unified simulation model foundation for subsequent clustering, filtering, and identification. It should be understood that the detailed model output by the model building module not only includes electrical parameters but also carries attribute data such as the rated capacity and geographical location of each photovoltaic unit for downstream modules to access.

[0063] The data acquisition module is used to perform power flow and transient calculations on the detailed model of the active distribution network, and to collect power and voltage values ​​during steady-state and transient periods at the equivalent boundary.

[0064] Specifically, the data acquisition module is signal-connected to the model building module and configured to receive detailed models and convert them into simulation calculation parameters. This module internally stores power flow calculation algorithms and transient simulation algorithms. During processing, the data acquisition module automatically records the active power, reactive power, and voltage amplitude during steady-state operation at equivalent boundary nodes, and acquires the time-series waveform data of these electrical quantities during transient periods after setting a three-phase short-circuit fault. Compared to general-purpose data acquisition units, this module is dedicated to extracting electrical quantities from equivalent boundary nodes, enabling it to directly map detailed simulation results into the input dataset required for subsequent parameter identification. This significantly reduces the computational cost of data preprocessing and ensures that the generated data naturally conforms to the input format of the parameter identification algorithm.

[0065] The clustering module is used to cluster distributed photovoltaic (PV) systems by using the low-voltage ride-through time of each PV system as a clustering characteristic index, and to determine the optimal number of clusters by using the silhouette coefficient.

[0066] Specifically, the clustering module is signal-connected to the model building module and is configured to receive the LVRT control parameters of each photovoltaic unit and extract its exit time point from low-voltage ride-through. This module includes a feature extraction unit and a clustering analysis unit. The feature extraction unit is configured to parse the active / reactive current response during the voltage dip and recovery periods from the LVRT control curves of each photovoltaic unit, thereby determining the timestamp of exiting the LVRT mode. The clustering analysis unit uses the AGNES hierarchical clustering algorithm, using the LVRT exit time as the unique feature vector to iteratively merge photovoltaic units, and evaluates the clustering quality under different cluster numbers through the silhouette coefficient, automatically selecting the cluster number with the largest silhouette coefficient as the optimal clustering result. This dual-unit parallel architecture ensures that the transient behavior differences of IBDGs are considered during the clustering process, while avoiding the subjectivity of manually setting the number of clusters through objective indicators.

[0067] The parameter filtering and aggregation module is used to extract undetermined parameters from the full set of electrical parameters to form a set of parameters to be filtered; based on the detailed model of the distribution network, the trajectory sensitivity of each parameter in the set of parameters to be filtered is calculated, and the parameters are filtered according to the trajectory sensitivity. The parameters that meet the conditions form a set of parameters to be identified, and the remaining parameters are directly aggregated through the clustering results.

[0068] Specifically, the parameter screening and aggregation module is signal-connected to the model building module and the clustering module, respectively, forming the core of the system's parameter dimensionality reduction. This module is configured to receive all electrical parameters from the model building module and screen out undetermined parameters involved in the equivalent model (such as induction motor stator and rotor resistance, reactance, inertial time constant, ZIP load factor, etc.) to form a set of parameters to be screened. In the trajectory sensitivity calculation branch, this module uses a detailed distribution network model to apply small perturbations to each parameter in the parameter set to be screened, observes the time-series changes of active and reactive power at the equivalent boundary, calculates the average trajectory sensitivity of each parameter to the two electrical quantities, and sums them; parameters whose sum of average trajectory sensitivities is greater than a preset threshold are identified as key parameters, forming a set of parameters to be identified. In the parameter aggregation branch, for non-key parameters, based on the clustering results output by the clustering module, this module uses the rated capacity weighted average method to aggregate the parameters of distributed photovoltaics within the same cluster to obtain the aggregated parameters of the equivalent unit. This dual-branch parallel architecture within the module enables the system to simultaneously filter key parameters and aggregate non-key parameters in the same data processing flow, without the need for additional manual intervention or secondary modeling.

[0069] The parameter identification module is used to construct a parameter identification objective function based on steady-state and transient time-series data at the iso-boundary. Based on the parameter identification objective function, a multi-objective improved marine predation algorithm is used to identify the parameters to be identified in the parameter set.

[0070] Specifically, the parameter identification module is signal-connected to both the data acquisition module and the parameter filtering and aggregation module, forming the core of the system's optimization solution. This module is configured to receive time-series data of the equivalent boundary from the data acquisition module and the set of parameters to be identified from the parameter filtering and aggregation module. In the objective function construction branch, this module uses the parameters to be identified as decision variables and minimizes the root mean square error between the measured values ​​of active and reactive power at the equivalent boundary and the output values ​​of the equivalent model as a dual objective, constructing a multi-objective optimization function. In the algorithm solution branch, this module employs an improved marine predator algorithm. This algorithm generates a Pareto front solution set through initializing the population with a good point set, adaptive CF parameter adjustment, and a three-stage update strategy. It then selects the compromise optimal solution from the Pareto front as the final identification result through fuzzy comprehensive evaluation. This multi-objective parallel optimization architecture enables the system to simultaneously consider the fitting accuracy of both active and reactive power sides, overcoming the problem of P and Q accuracy imbalance in traditional single-objective identification.

[0071] The equivalent modeling module is used to construct an equivalent model of multiple machines in an active distribution network based on the clustering results and parameter identification results.

[0072] Specifically, the equivalent modeling module is signal-connected to both the clustering module and the parameter identification module. It is configured to receive the clustering results and identified parameter values, and assemble the equivalent units corresponding to each cluster according to their topological connections in the distribution network, forming a complete active distribution network equivalent model containing multiple equivalent units. This equivalent model can be directly exported as a standard format simulation file for subsequent large-scale power grid transient stability analysis.

[0073] Through the coordinated configuration and data interaction of the above modules, the system provided in this embodiment solidifies the complex equivalent modeling method into a stable device function. The modules communicate with each other through standardized data interfaces, ensuring both focus on individual functions and seamless integration of the overall process. It should be understood that although this embodiment is described in the form of functional modules, in actual hardware implementation, these modules can be carried by independent processors, FPGAs, ASICs, or software units integrated into the same controller. As long as they can realize the above data flow and control logic, they all fall within the protection scope of this application.

[0074] Example 3 This embodiment provides a terminal and a computer-readable storage medium as the hardware carrier and program product for implementing the aforementioned active power distribution network equivalent modeling method based on multi-objective parameter identification. The terminal and storage medium correspond one-to-one with the steps described in Embodiment 1 in technical substance, but in terms of legal protection, they respectively point to specific electronic devices and software products, ensuring that the technical solution of this application can cover the entire lifecycle protection needs from research and development, manufacturing to sales and use.

[0075] The terminal provided in this application includes a processor and a storage medium.

[0076] The storage medium is used to store instructions.

[0077] The processor is used to operate according to instructions to perform the steps of the active distribution network equivalent modeling method based on multi-objective parameter identification as described in Example 1.

[0078] Specifically, this terminal constitutes the core computing unit of the active distribution network equivalent modeling system. The processor communicates with the storage medium via a system bus or dedicated interface, enabling the processor to read and parse the pre-stored program instructions in the storage medium. In actual deployment, the processor can be any one or a combination of a general-purpose microprocessor, digital signal processor, application-specific integrated circuit (ASIC), or field-programmable gate array (FPGA), as long as its computing power meets the timing requirements of large-scale distribution network simulation and multi-objective optimization solutions. The storage medium includes, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, or electrically erasable programmable read-only memory (EEPROM), used to persistently or temporarily store the machine code or scripts required to implement steps one through five in Embodiment 1. When the terminal is powered on, the processor loads these instructions, thereby driving the externally connected simulation server or data acquisition equipment to complete a series of operations such as model building, power flow calculation, clustering, trajectory sensitivity screening, multi-objective MPA identification, and equivalent modeling. This limitation on the hardware carrier means that the scope of protection of this application is not limited to abstract control logic, but extends to physical devices such as specific industrial control computers, embedded controllers or edge computing nodes that carry the logic, providing a clear legal basis for subsequent infringement comparisons of hardware products.

[0079] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the active distribution network equivalent modeling method based on multi-objective parameter identification as described in Embodiment 1.

[0080] Specifically, the computer-readable storage medium is a tangible carrier of software technology. It should be understood that the term "computer-readable storage medium" as used herein refers to a tangible device capable of holding and storing instructions used by an instruction execution device, explicitly excluding transient signals such as radio waves, light pulses, or electrical signals transmitted via wires. As a preferred embodiment, the storage medium may be a portable computer disk, hard disk, optical disk, digital multifunction disk, memory stick, or semiconductor storage device, etc. The computer program stored in this medium contains all the logical code implementing the method described in Embodiment 1, such as a feature extraction module for calculating LVRT exit time, a hierarchical merging module for performing AGNES clustering, a perturbation simulation module for calculating trajectory sensitivity, a population evolution module for performing improved MPA multi-objective optimization, and a parameter assembly module for generating an equivalent model, etc. When the storage medium is installed or connected to any compatible computing device, and the device's processor calls and executes the program therein, the computing device possesses the active power distribution network equivalent modeling capability described in this application. By solidifying the technical solution into an independent storage medium product, this application can obtain effective intellectual property protection in various commercial scenarios such as software distribution, system integration, or cloud deployment. Even if the equivalent modeling algorithm is copied to other brands' hardware platforms, as long as it calls the program code stored in the medium, it may still fall within the protection scope of this application.

[0081] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0082] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0083] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0084] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0085] Finally, 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 the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. An equivalent modeling method for active distribution networks based on multi-objective parameter identification, characterized in that, include: Based on the active distribution network, the equivalent boundary is determined, all electrical parameters are collected, and a detailed model of the distribution network including distributed photovoltaic power is constructed. Power flow and transient calculations are performed on a detailed model of an active distribution network, and power and voltage values ​​during steady-state and transient periods are collected at equivalent boundaries. The distributed photovoltaic system is grouped using the low-voltage ride-through time of each distributed photovoltaic system as a clustering characteristic index, and the optimal number of clusters is determined by the silhouette coefficient. Undetermined parameters are extracted from the full set of electrical parameters to form a set of parameters to be screened. Based on the detailed model of the distribution network, the trajectory sensitivity of each parameter in the parameter set to be screened is calculated. The parameters are screened according to the trajectory sensitivity, and the parameters that meet the conditions constitute the parameter set to be identified. The remaining parameters are directly aggregated through the clustering results. Based on steady-state and transient time-series data at equivalent boundaries, a parameter identification objective function is constructed. Based on the objective function of parameter identification, a multi-objective improved marine predation algorithm is used to identify the parameters to be identified in the parameter set; based on the clustering results and parameter identification results, an equivalent model of multiple machines in an active power distribution network is constructed.

2. The method for equivalent modeling of active distribution networks based on multi-objective parameter identification according to claim 1, characterized in that: Using the low-voltage ride-through time of each distributed photovoltaic (PV) photovoltaic (PV) system as a clustering characteristic index, distributed PV systems are clustered. The optimal number of clusters is determined by the silhouette coefficient, including: The moment when the terminal voltage of each distributed photovoltaic system recovers to a preset voltage threshold during low-voltage ride-through is obtained, and this moment is used as the exit time of low-voltage ride-through to construct a cluster feature vector; A hierarchical clustering algorithm is used to calculate the Euclidean distance between samples and the distance between clusters based on the cluster feature vectors. The clustering tree is generated by iteratively merging the clusters with the smallest distance. Calculate the overall profile coefficient corresponding to different numbers of clusters in the clustering tree, determine the number of clusters corresponding to the largest overall profile coefficient as the optimal number of clusters, and divide each distributed photovoltaic power generation into multiple clusters based on the optimal number of clusters.

3. The method for equivalent modeling of active distribution networks based on multi-objective parameter identification according to claim 2, characterized in that: The calculation of Euclidean distance between samples and inter-cluster distance based on the cluster feature vectors includes: The absolute value of the difference between the exit low-voltage ride-through times of any two distributed photovoltaic systems is taken as the Euclidean distance between the samples. The inter-cluster distance is calculated by dividing the product of the number of samples in any two clusters by the sum of the number of samples in those two clusters, and then multiplying by the square of the norm of the difference between the means of samples in those two clusters. The calculation of the overall profile coefficient corresponding to different numbers of clusters in the clustering tree includes: For each sample in the clustering tree, the average distance from the sample to other samples in its own cluster is calculated as the first distance, and the average distance from the sample to all samples in the nearest other cluster is calculated as the second distance; The difference between the second distance and the first distance is divided by the maximum value of the first distance and the second distance to obtain the contour coefficient of the sample; The average of the silhouette coefficients of all samples is taken as the overall silhouette coefficient.

4. The method for equivalent modeling of active distribution networks based on multi-objective parameter identification according to claim 1, characterized in that: The trajectory sensitivity is calculated for each parameter in the parameter set to be screened. Parameters are then screened based on the trajectory sensitivity, and parameters that meet the criteria form the parameter set to be identified. The remaining parameters are directly aggregated using the clustering results, including: Calculate the average trajectory sensitivity of each parameter in the parameter set to be screened to multiple electrical quantities of the active distribution network, and sum the average trajectory sensitivity of each electrical quantity; The parameters whose sum of average trajectory sensitivities is greater than a preset sensitivity threshold are determined as undetermined parameters, forming the set of parameters to be identified; For the remaining parameters, based on the clustering results, the parameters of distributed photovoltaics within the same cluster are aggregated using the rated capacity weighted average method to obtain the aggregated parameters of the equivalent unit.

5. The method for equivalent modeling of active distribution networks based on multi-objective parameter identification according to claim 4, characterized in that: Calculate the average trajectory sensitivity of each parameter in the parameter set to be screened to multiple electrical quantities in the active distribution network, and sum the average trajectory sensitivity of each electrical quantity, including: For any parameter in the set of parameters to be screened, calculate the instantaneous trajectory sensitivity of the parameter at each sampling time within a preset observation period. The instantaneous trajectory sensitivity is the ratio of the relative change rate of each electrical quantity to the relative change rate of the parameter when the parameter changes. Integrate the instantaneous trajectory sensitivity within the preset observation period and calculate the average value to obtain the average trajectory sensitivity of the parameter for each electrical quantity; The method of aggregating parameters of distributed photovoltaic systems within the same cluster using the rated capacity weighted average method to obtain aggregated parameters for equivalent units includes: The aggregate parameter of the equivalent unit is obtained by summing the products of the rated capacity of each distributed photovoltaic unit within the same cluster and the corresponding parameter value, and then dividing by the sum of the rated capacities of each distributed photovoltaic unit within the same cluster.

6. The method for equivalent modeling of active distribution networks based on multi-objective parameter identification according to claim 1, characterized in that: The parameter identification objective function is constructed based on the steady-state and transient time-series data at the equivalent boundary, including: The parameter identification objective function is constructed by taking the root mean square error of active power and the root mean square error of reactive power at the grid connection point of the active distribution network as two independent objective terms. The root mean square error of active power is the root mean square of the sum of squares of the differences between the reference value of active power at the equivalent boundary and the calculated value of the equivalent model. The root mean square error of reactive power is the root mean square of the sum of squares of the differences between the reactive power reference value and the calculated value of the equivalent model at the equivalent boundary.

7. The method for equivalent modeling of active distribution networks based on multi-objective parameter identification according to claim 1, characterized in that: An improved multi-objective marine predation algorithm is used to identify parameters in the parameter set to be identified, including: An initial population uniformly covering the solution space is generated using the optimal point set theory. A fast non-dominated sort is performed on the initial population. The initial population is divided into multiple non-dominated layers according to the dominance relationship. Individuals are selected from the first non-dominated layer to construct an elite matrix. An adaptive step size control factor is introduced, and the optimization process is divided into multiple stages according to the current iteration number. Different motion strategies are used in different stages to update the position of individuals in the population until the maximum number of iterations is reached, and the final Pareto optimal parameter set is output.

8. The method for equivalent modeling of active distribution networks based on multi-objective parameter identification according to claim 7, characterized in that: The introduction of an adaptive step size control factor divides the optimization process into multiple stages based on the current iteration number, employing different movement strategies to update the individual positions in the population at different stages, including: The adaptive step size control factor is calculated based on the current iteration number, the maximum iteration number, and the preset initial and final step size values. When the current iteration count is less than one-third of the maximum iteration count, the prey position is updated using a Brownian motion random vector for global exploration. When the current iteration number is greater than or equal to one-third of the maximum iteration number and less than two-thirds of the maximum iteration number, the first half of the population is explored using a Levy flight random vector, and the second half of the population is developed using a Brownian motion random vector. When the current iteration number is greater than or equal to two-thirds of the maximum iteration number, the Levy flight random vector is used to develop all populations to approach the global optimum region.

9. The method for equivalent modeling of active distribution networks based on multi-objective parameter identification according to claim 1, characterized in that: After the multi-objective improved marine predation algorithm identifies the parameters to be identified in the parameter set, the method further includes: Obtain the maximum and minimum root mean square error of each electrical quantity in the Pareto optimal parameter set of the parameter identification output; For any set of candidate parameters in the Pareto optimal parameter set, the difference between the maximum value of the root mean square error of the electrical quantity and the function value of the candidate parameter on the root mean square error of the electrical quantity is divided by the difference between the maximum value and the minimum value to obtain the linear membership degree of the candidate parameter on the root mean square error of the electrical quantity. The average of the linear membership degree of the candidate parameter on the root mean square error of active power and the linear membership degree on the root mean square error of reactive power is used to obtain the overall satisfaction of the candidate parameter. The candidate parameter with the highest overall satisfaction is selected as the optimal compromise parameter. The optimal compromise parameter and the remaining parameters after parameter aggregation through the clustering results are used to generate the multi-machine equivalent model of the active distribution network.

10. An active distribution network equivalent modeling system based on multi-objective parameter identification using the method described in any one of claims 1-8, comprising a model building module, a data acquisition module, a clustering module, a parameter filtering and aggregation module, a parameter identification module, and an equivalent modeling module, characterized in that: The model building module is used to determine the equivalent boundary based on the active distribution network, collect all electrical parameters, and build a detailed model of the distribution network including distributed photovoltaics; The data acquisition module is used to perform power flow and transient calculations on the detailed model of the active distribution network, and to collect power and voltage values ​​during steady-state and transient periods at the equivalent boundary. The clustering module is used to cluster distributed photovoltaics using the exit low-voltage ride-through time of each distributed photovoltaic unit as a clustering characteristic index, and to determine the optimal number of clusters through the silhouette coefficient. The parameter filtering and aggregation module is used to extract undetermined parameters from the full set of electrical parameters to form a set of parameters to be filtered; calculate the trajectory sensitivity of each parameter in the set of parameters to be filtered, filter the parameters according to the trajectory sensitivity, form a set of parameters to be identified that meet the conditions, and aggregate the remaining parameters directly through the clustering results. The parameter identification module is used to construct a parameter identification objective function based on steady-state and transient time-series data at the equivalent boundary. Based on the parameter identification objective function, a multi-objective improved marine predation algorithm is used to identify the parameters to be identified in the parameter set. The equivalent modeling module is used to construct an equivalent model of multiple machines in an active distribution network based on the clustering results and parameter identification results.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1-9.

12. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-9.

Images