Improved artificial bee algorithm-based active power distribution network dynamic cluster partitioning method, device and medium

Abstract

The application discloses a kind of active distribution network dynamic cluster partitioning method, equipment and medium based on improved artificial bee algorithm, method includes: based on active distribution network data, modulus index, voltage regulation capability index and load fluctuation suppression capability index are established, for constructing time-varying comprehensive evaluation function;With candidate cluster partitioning scheme as the individual position of improved artificial bee algorithm, with time-varying comprehensive evaluation function as fitness basis, search the candidate cluster partitioning scheme of current operation period using improved artificial bee algorithm;The new partitioning scheme obtained by re-searching the current operation period is compared with the maintenance scheme of the cluster structure of the last period, and the cluster partitioning result finally used in the current operation period is determined according to the comparison result of evaluation gain and preset update threshold value.The application improves the adaptability, stability and engineering implementability of the partitioning result.

Description

Technical Field

[0001] This invention relates to the field of active distribution network optimization control technology, specifically to an active distribution network dynamic cluster partitioning method, device, and medium based on an improved artificial hummingbird algorithm. Background Technology

[0002] With the integration of various resources such as distributed photovoltaic (PV) power, energy storage devices, static var compensators (SVCs), parallel capacitor banks, and controllable loads into the distribution network, the traditional distribution network is gradually evolving from a unidirectional power supply network into an active distribution network with multi-source coordination among power sources, grids, loads, and storage. While a high proportion of distributed PV power can improve the utilization of clean energy, its output is random, fluctuating, and time-varying, which can easily lead to problems such as increased node voltage deviation, local voltage exceeding limits, changes in power flow direction, and complex coordination and control of regulation resources.

[0003] Existing active distribution network partitioning or clustering methods typically involve static partitioning based on network topology, electrical distance, geographical location, or typical operating scenarios. While these methods can reduce the scale of optimization control to some extent, they struggle to reflect changes in node coupling relationships caused by variations in daily load levels, distributed photovoltaic output, and regulatory resource operating margins, leading to discrepancies between cluster boundaries and actual voltage regulation requirements.

[0004] While some existing methods consider modularity or electrical coupling strength, they do not adequately address the reactive power-voltage response characteristics of nodes, differences in resource regulation capabilities, and the cluster's ability to suppress net load fluctuations. This can easily lead to problems such as a compact cluster structure but insufficient regulation capacity, or an uneven distribution of regulation resources. Technical solutions using intelligent optimization algorithms to solve cluster partitioning also generally suffer from issues such as low initial population quality in dynamic time-series scenarios, weakened search capabilities in later stages, susceptibility to local optima, and frequent adjustments to cluster boundaries.

[0005] Therefore, it is necessary to propose an active distribution network cluster partitioning method that can simultaneously consider the static network structure, node voltage response characteristics, resource function characteristics, and dynamic changes in operating status. Furthermore, it is necessary to improve the adaptability, stability, and engineering feasibility of the partitioning results by improving intelligent optimization algorithms and dynamic update mechanisms. Summary of the Invention

[0006] This invention addresses the shortcomings of existing active distribution network cluster partitioning methods, such as insufficient adaptability of static partitioning, difficulty in balancing structural and functional aspects with a single index, susceptibility of intelligent optimization algorithms to local optima, and frequent adjustments to cluster boundaries. It provides a dynamic active distribution network cluster partitioning method, equipment, and medium based on an improved artificial hummingbird algorithm. This invention constructs a node comprehensive similarity matrix by building static similarity of electrical distance, dynamic similarity of reactive power-voltage sensitivity, and dynamic similarity of resource characteristics. Furthermore, it establishes an evaluation system for modularity, voltage regulation capability, and load fluctuation suppression capability, and utilizes the improved artificial hummingbird algorithm and threshold triggering mechanism to achieve dynamic updates of the cluster structure, thereby improving the adaptability of the cluster partitioning results to time-varying operating scenarios.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] An active distribution network dynamic cluster partitioning method based on an improved artificial hummingbird algorithm includes the following steps: Based on active distribution network data, modularity index, voltage regulation capability index, and load fluctuation suppression capability index are established to construct a time-varying comprehensive evaluation function. The candidate cluster partitioning scheme is used as the individual position of the improved artificial hummingbird algorithm, and the time-varying comprehensive evaluation function is used as the fitness criterion. The improved artificial hummingbird algorithm is used to search for candidate cluster partitioning schemes in the current runtime. The new partitioning scheme obtained by the current runtime segment is evaluated and compared with the scheme that maintains the cluster structure of the previous time segment. The final cluster partitioning result adopted by the current runtime segment is determined based on the comparison result of the evaluation gain and the preset update threshold. The cluster partitioning result is output, which includes node cluster labels, cluster boundary information and branch cutting information, and is used for active distribution network zonal optimization scheduling, intraday rolling optimization or inter-cluster coordination control.

[0009] To optimize the above technical solution, the specific measures also include: Furthermore, the establishment of modularity index, voltage regulation capability index, and load fluctuation suppression capability index based on active distribution network data specifically includes: Calculate the electrical path impedance between any two nodes based on topology data and line parameters, and construct a node static similarity matrix based on electrical distance; Based on the power flow calculation results of the current operating period, a power flow Jacobian matrix is ​​established, the approximate sensitivity relationship between reactive power injection changes and node voltage changes is extracted, and a node dynamic similarity matrix based on reactive power-voltage sensitivity is constructed. Based on the parameters of distributed photovoltaics, photovoltaic inverters, static var compensators, parallel capacitor banks, and the configuration and operational margin of energy storage devices at the nodes, a node resource feature vector is constructed, and a dynamic similarity matrix of resource features is constructed from the node resource feature vector. The node static similarity matrix, the node dynamic similarity matrix based on reactive power-voltage sensitivity, and the resource feature dynamic similarity matrix are weighted and fused to obtain the node comprehensive similarity matrix for the current operating segment. ; Based on the node comprehensive similarity matrix, a modularity index is established; The voltage regulation capability index is calculated by combining node voltage deviation, sensitivity relationship, and intra-cluster regulation resource margin. The load fluctuation suppression capability index is calculated by combining the cluster net load fluctuation and the active power regulation resource capacity.

[0010] Furthermore, the process of constructing the node static similarity matrix is ​​as follows: node i With nodes j The set of branches on the path between them is Ω. ij branch road ( m , n The impedance is Z mn Then the node i With nodes j Total impedance of the path between and electrical distance They are respectively:

[0011]

[0012] The electrical distance is mapped to node static similarity using a Gaussian kernel function:

[0013] In the formula, For electrical distance scale parameters, It is a node i With nodes j Static similarity between nodes; traverse all nodes, and the static similarity between all nodes forms a node static similarity matrix. .

[0014] Furthermore, the process of constructing the node dynamic similarity matrix is ​​as follows: Perform power flow calculations during the current runtime to obtain the power flow Jacobian matrix. :

[0015] In the formula, H This represents the sensitivity matrix of nodal active power injection to changes in voltage phase angle. N This represents the sensitivity matrix of nodal active power injection to changes in voltage amplitude. K This represents the sensitivity matrix of nodal reactive power injection to changes in voltage phase angle. L This represents the sensitivity matrix of nodal reactive power injection to changes in voltage amplitude; Based on the linearized power flow relationship, the node voltage increment U Changes in reactive power injection at nodes Q Connecting these points, we get:

[0016]

[0017] In the formula, This is an approximate reactive power-voltage sensitivity matrix. elements in Represents a node j Changes in reactive power injection on nodes i The degree of influence of voltage changes; Based on the approximate reactive power-voltage sensitivity matrix, the time interval is defined. t Next node i With nodes j Sensitivity difference index between for:

[0018] In the formula, Represents a node i The degree to which changes in reactive power injection affect the voltage itself. Represents a node j The degree to which changes in reactive power injection affect the voltage itself. Represents a node j Reactive power injection changes on nodes i The degree of influence of voltage changes Represents a node i Reactive power injection changes on nodes j The degree of influence of voltage changes; Dynamic node similarity is constructed using a Gaussian kernel function:

[0019] In the formula, s qv This is a parameter for adjusting the reactive power-voltage sensitivity similarity. For nodes i With nodes jThe dynamic similarity is calculated by traversing all nodes, and the dynamic similarity of all nodes forms a node dynamic similarity matrix. .

[0020] Furthermore, the specific steps of constructing node resource feature vectors and building a dynamic similarity matrix of resource features from these node resource feature vectors are as follows: Define nodes i At any moment t resource feature vector for:

[0021] In the formula, Represents a node i Contributing to photovoltaic power generation; Indicates the photovoltaic inverter at time t Maximum adjustable reactive power capacity; Represents a node i The rated apparent capacity of the photovoltaic inverter; Indicates that the photovoltaic inverter is at the node i time t Those who have made meritorious contributions; Represents a node i The reactive power adjustable quantity of the SVC; Represents a node i The maximum reactive power compensation capacity of the parallel capacitor bank, of which The maximum number of groups that can be added. For single-group compensation capacity; and Representing nodes respectively i The maximum charging and discharging power and rated capacity of the energy storage device are specified. If the node is not configured with the corresponding type of resource, the corresponding component is set to 0. After normalizing the features of each dimension, it is defined at time... t Next node i With nodes j Differences in resource characteristics between for:

[0022] In the formula, and Representing nodes respectively i With nodes j At any moment t Normalized resource feature vector; Represents the L2 norm; time t Next node i With nodes j Resource feature similarity between for:

[0023] In the formula, As the resource feature similarity adjustment parameter, we iterate through all nodes, and the resource feature similarity of all nodes forms a dynamic resource feature similarity matrix. .

[0024] Furthermore, the establishment of the modularity index based on the node comprehensive similarity matrix is ​​specifically as follows: Time t The node comprehensive similarity matrix below Viewed as a weighted undirected network, it is called a comprehensive similarity network, where the network nodes are distribution network nodes participating in cluster partitioning, and the edge weights are... Represents a node i With nodes j The overall similarity between them defines the nodes. i Weighting degree in the network for:

[0025] The total edge weight of the network is then expressed as:

[0026] In the formula, N This represents the total number of nodes participating in the cluster partitioning; express t At any given moment i The weighting degree; m t Indicates in t The total edge weight in a comprehensive similar network at any given time; Set time t Next node i Cluster number c i,t If node i With nodes j If they are in the same cluster, then the cluster similarity / difference identifier is used. d ( c i,t , c j,t )=1, otherwise d ( c i,t , c j,t )=0; the original modularity corresponding to the partitioning scheme. Represented as:

[0027] In the formula, expresst At any given moment j The weighted degree is used to normalize the original modularity index to obtain the normalized modularity index. .

[0028] Furthermore, the calculated voltage regulation capability index specifically includes: set up t Time of the first c The set of nodes in each cluster is Oh c,t The maximum voltage deviation within this cluster is expressed as: , will the c In each cluster t The node with the largest absolute value of voltage deviation at any given time is defined as the critical node. The maximum voltage correction capability of resources within the cluster for this critical node. Represented as:

[0029] In the formula, express t Time Node j Active power regulation for key nodes Approximate sensitivity to voltage changes express t Time Node j Reactive power regulation for critical nodes Sensitivity to voltage changes; and They are respectively t Time Node The active and reactive power regulation capabilities that can be provided; No. c Each cluster t Voltage regulation capability index at any time Defined as The system as a whole is t Voltage regulation capability index at any time Take the average value of each cluster; The specific index for calculating load fluctuation suppression capability is as follows: No. c Each cluster during the time period t Net load fluctuation Represented as:

[0030] In the formula, and They are nodes i During the period t Active load and photovoltaic active power output; c Each cluster during the time period t The active power regulation capability that can be used to smooth out fluctuations is , No. c Each cluster within the window period t Load fluctuation suppression capability Represented as The system is at all times t Overall load fluctuation suppression capability index Defined as In the formula, K t for t The total number of clusters corresponding to the current partitioning scheme at any given time; At the current moment t A benchmark load fluctuation evaluation window.

[0031] Furthermore, the time-varying comprehensive evaluation function is specifically as follows:

[0032] in, It is a time-varying comprehensive evaluation function. oh 1. oh 2 and oh 3 represents the evaluation weight. It is a normalized modularity metric. Is the system as a whole in t Voltage regulation capability index at any given time. Is the system at any time t The overall load fluctuation suppression capability index; The process of using the improved artificial hummingbird algorithm to search for candidate cluster partitioning schemes in the current runtime segment specifically involves: Step a: Initialize algorithm parameters; randomly generate a number of hummingbird individuals in the solution space as the initial population; Step b: The best individual from the previous time period is superimposed with random perturbation and used as part of the initial population for the current time period; Step c: Update the position of each individual hummingbird according to the following formula:

[0033] Where, η i For random mutation vectors, p m ( k ) is the first k The mutation probability is adjusted based on the number of iterations. K This represents the maximum number of iterations. It is the first k+ 1st generation i The location of an individual hummingbird It is the lower bound of the mutation probability. It is the upper limit of the mutation probability; For boundary projection operators; Step d: After the location is updated, the candidate cluster partitioning scheme is discretized and its feasibility is corrected to meet the requirements of cluster number range, minimum cluster node number and network connectivity. Step e: Based on the comprehensive evaluation function c t Calculate the fitness value of each hummingbird, and select the one with the best fitness among all hummingbirds as the current optimal solution; Step f: Check if the termination condition is met. If not, directly retain one or more individuals with the best current fitness to the next generation and return to step c. If yes, output the current optimal solution as the cluster partitioning result.

[0034] The present invention also proposes an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the active distribution network dynamic cluster partitioning method based on the improved artificial hummingbird algorithm as described above.

[0035] The present invention also proposes a computer-readable storage medium storing a computer program that enables a computer to execute the active distribution network dynamic cluster partitioning method based on the improved artificial hummingbird algorithm described above.

[0036] The beneficial effects of this invention are: (1) The present invention constructs a comprehensive similarity matrix from three dimensions: electrical distance, reactive power-voltage sensitivity and resource characteristics. Compared with the division method based solely on topology or single electrical distance, it can more comprehensively characterize the structural coupling, dynamic response and resource function association between nodes.

[0037] (2) The present invention establishes an evaluation system consisting of modularity, voltage regulation capability and load fluctuation suppression capability, so that the cluster division result not only meets the structural requirements of strong coupling within the cluster and weak coupling between the clusters, but also takes into account voltage support capability and source load fluctuation suppression capability.

[0038] (3) The present invention introduces the initialization of the optimal solution of the previous time period, the elite retention and the adaptive mutation mechanism into the artificial hummingbird algorithm, which improves the initial population quality, global search ability and local development ability in dynamic time series scenarios, and reduces the possibility of getting trapped in local optima.

[0039] (4) This invention determines whether to update the cluster structure by evaluating the gain and the update threshold, which can achieve a balance between dynamic adaptability and structural stability, avoid frequent adjustments to the cluster boundary, and improve the feasibility of the project.

[0040] (5) The cluster labels, boundary information and branch cutting information output by this invention can be directly used as the basis for intraday rolling optimization, inter-group coordinated control and distributed optimization modeling of active distribution network, thereby improving the scalability and real-time performance of subsequent control models. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the active distribution network dynamic cluster partitioning method based on the improved artificial hummingbird algorithm proposed in this invention. Figure 2 This is a diagram showing the dynamic partitioning results of the active distribution network cluster obtained using the method of this invention. Figure 3 A comparison chart of voltage regulation capability indicators under different cluster partitioning schemes; Figure 4 A comparison chart of load fluctuation suppression capabilities under different cluster partitioning schemes. Detailed Implementation

[0042] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0043] Example 1 This invention proposes an active distribution network dynamic cluster partitioning method based on an improved artificial hummingbird algorithm. The process of this method is as follows: Figure 1 As shown, it includes the following steps: Step 1: Collect data on the active distribution network topology, branch impedance, node load, distributed photovoltaic output, photovoltaic inverter capacity, static var compensator (SVC) capacity, capacitor bank (CB) compensation capacity, maximum charging and discharging power and rated capacity of energy storage system (ESS), and node voltage operation data. For each operating period, update node characteristics and power flow operation status based on the current load level, photovoltaic output, and the status of various regulating resources.

[0044] Step 2: Based on active distribution network data, establish modularity index, voltage regulation capability index, and load fluctuation suppression capability index to construct a time-varying comprehensive evaluation function; Step 2.1: Calculate the electrical path impedance between any two nodes based on the topology data and line parameters, and construct a node static similarity matrix based on electrical distance. The process of constructing the node static similarity matrix is ​​as follows: For a radial active distribution network, there is a unique electrical path between any two nodes. i With nodes jThe set of branches on the path between them is Ω. ij branch road ( m , n The impedance is Z mn Then the node i With nodes j Total impedance of the path between and electrical distance They are respectively:

[0045]

[0046] The electrical distance is mapped to node static similarity using a Gaussian kernel function:

[0047] In the formula, The electrical distance scale parameter can be the median of all non-zero node electrical distance samples. The smaller the electrical distance, the greater the static similarity, indicating a stronger inherent electrical coupling between the two nodes at the network topology and line parameter levels. It is a node i With nodes j Static similarity between nodes; traverse all nodes, and the static similarity between all nodes forms a node static similarity matrix. .

[0048] Step 2.2: Based on the power flow calculation results of the current operating period, establish the power flow Jacobian matrix, extract the approximate sensitivity relationship between reactive power injection changes and node voltage changes, and construct a node dynamic similarity matrix based on reactive power-voltage sensitivity. The construction process of the node dynamic similarity matrix is ​​as follows: Perform power flow calculations during the current runtime to obtain the power flow Jacobian matrix. :

[0049] In the formula, H This represents the sensitivity matrix of nodal active power injection to changes in voltage phase angle. N This represents the sensitivity matrix of nodal active power injection to changes in voltage amplitude. K This represents the sensitivity matrix of nodal reactive power injection to changes in voltage phase angle. L This represents the sensitivity matrix of nodal reactive power injection to changes in voltage amplitude; Based on the linearized power flow relationship, the node voltage increment can be... U Changes in active and reactive power injection at nodes P , QConnecting them, that is:

[0050] The main focus is on the impact of reactive power injection on node voltage, and the following results are obtained:

[0051]

[0052] In the formula, This is an approximate reactive power-voltage sensitivity matrix, used to characterize the impact of changes in reactive power injection at nodes on voltage changes at each node. elements in Represents a node j Changes in reactive power injection on nodes i The degree of influence of voltage changes.

[0053] Based on the approximate reactive power-voltage sensitivity matrix, the time interval is defined. t Next node i With nodes j Sensitivity difference index between for:

[0054] In the formula, Represents a node i The degree to which changes in reactive power injection affect the voltage itself. Represents a node j The degree to which changes in reactive power injection affect the voltage itself. Represents a node j Reactive power injection changes on nodes i The degree of influence of voltage changes Represents a node i Reactive power injection changes on nodes j The degree of influence of voltage changes; Dynamic node similarity is constructed using a Gaussian kernel function:

[0055] In the formula, s qv This is the reactive power-voltage sensitivity similarity adjustment parameter. The larger this parameter is, the weaker the attenuation effect of the sensitivity difference between different nodes. For nodes i With nodes j The dynamic similarity, when D qv,t ( i , j The larger the value, the stronger the node. i With nodes jThese nodes exhibit more similar voltage response characteristics under reactive power disturbances, making them more suitable for grouping into the same cluster. The dynamic similarity of all nodes, obtained by traversing all nodes, forms a node dynamic similarity matrix. .

[0056] Step 2.3: Based on the parameters of distributed photovoltaic power generation, photovoltaic inverters, static var compensators, parallel capacitor banks, and the configuration and operational margin of energy storage devices at the nodes, construct node resource feature vectors, and then construct a dynamic similarity matrix of resource features from these node resource feature vectors; specifically: To reflect the differences in adjustable resource allocation and adjustment capabilities at nodes, nodes are defined. i At any moment t resource feature vector for:

[0057] In the formula, Represents a node i Contributing to photovoltaic power generation; Indicates the photovoltaic inverter at time t Maximum adjustable reactive power capacity; Represents a node i The rated apparent capacity of the photovoltaic inverter; Indicates that the photovoltaic inverter is at the node i time t Those who have made meritorious contributions; Represents a node i The reactive power adjustable quantity of the SVC; Represents a node i The maximum reactive power compensation capacity of the parallel capacitor bank, of which The maximum number of groups that can be added. For single-group compensation capacity; and Representing nodes respectively i The maximum charging and discharging power and rated capacity of the energy storage device are specified. If the node is not configured with the corresponding type of resource, the corresponding component is set to 0. After normalizing the features of each dimension, it is defined at time... t Next node i With nodes j Differences in resource characteristics between for:

[0058] In the formula, and Representing nodes respectively i With nodes j At any moment t Normalized resource feature vector; Denotes the 2-norm; if e r,t ( i , j The smaller the value, the closer the two nodes are in terms of resource allocation and adjustment capabilities.

[0059] time t Next node i With nodes j Resource feature similarity between for:

[0060] In the formula, As the resource feature similarity adjustment parameter, we iterate through all nodes, and the resource feature similarity of all nodes forms a dynamic resource feature similarity matrix. .when e r,t ( i , j The smaller the time, the more accurate the time measurement. t Next node i With nodes j The closer the similarity in resource allocation and regulation capabilities of photovoltaic inverters, SVCs, CBs, and energy storage, the higher the resource feature similarity obtained by the Gaussian kernel function. D r,t ( i , j The larger the number of nodes with similar resource characteristics, the stronger the consistency in their regulation functions and response characteristics. Assigning them to the same cluster helps reduce control differences within the cluster and facilitates subsequent local coordination and optimization.

[0061] Step 2.4: Perform weighted fusion of the node static similarity matrix, the node dynamic similarity matrix based on reactive power-voltage sensitivity, and the resource feature dynamic similarity matrix to obtain the node comprehensive similarity matrix for the current operating segment. ;

[0062] in, l e , l qv and l r These are three types of similarity weights, satisfying... l e + l qv + l r =1. Comprehensive similarity matrix It is used to uniformly characterize the structural association, dynamic voltage response association, and resource function association between nodes.

[0063] Step 2.5: Based on the node comprehensive similarity matrix, establish a modularity index; specifically: Time t The node comprehensive similarity matrix below Viewed as a weighted undirected network, it is called a comprehensive similarity network, where the network nodes are distribution network nodes participating in cluster partitioning, and the edge weights are... Represents a node i With nodes j The overall similarity between them defines the nodes. i Weighting degree in the network for:

[0064] The total edge weight of the network is then expressed as:

[0065] In the formula, N This represents the total number of nodes participating in the cluster partitioning; express t At any given moment i The weighting degree; m t Indicates in t The total edge weight in a comprehensive similar network at any given time; Set time t Next node i Cluster number c i,t If node i With nodes j If they are in the same cluster, then the cluster similarity / difference identifier is used. d ( c i,t , c j,t )=1, otherwise d ( c i,t , c j,t )=0; the original modularity corresponding to the partitioning scheme. Represented as:

[0066] In the formula, express t At any given moment j The weighting degree, M raw,t The larger the value, the closer the connections between nodes within the cluster and the weaker the connections between clusters at the current moment, resulting in a better structured partition. This takes into account different operating scenarios. M raw,tThe value range may vary. To avoid inconsistencies with other functional indicators, the original modularity index is normalized to obtain a normalized modularity index. .

[0067]

[0068] In the formula, M min and M max These represent the lower and upper bounds of the sample modularity, respectively. (Normalized) Q,t The larger the value, the more reasonable the current time period division scheme is in terms of structure.

[0069] Step 2.6: Calculate the voltage regulation capability index by combining node voltage deviation, sensitivity relationship, and cluster regulation resource margin; specifically: set up t Time of the first c The set of nodes in each cluster is Oh c,t The maximum voltage deviation within this cluster is expressed as: :

[0070] In the formula, V i,t for t Time Node i voltage amplitude, V ref For the rated voltage, this embodiment takes... V ref =1.0pu.

[0071] To measure the voltage regulation requirements of the weakest node in the cluster, the first... c In each cluster t The node with the largest absolute value of voltage deviation at any given time is defined as the critical node. ,Right now:

[0072] The critical node exhibits the largest voltage deviation, best reflecting the current voltage regulation needs of the cluster. If multiple nodes share the same maximum voltage deviation, any one of them can be selected as the critical node.

[0073] Maximum voltage correction capability of resources within the cluster for this critical node Represented as:

[0074] In the formula, express tTime Node j Active power regulation for key nodes Approximate sensitivity to voltage changes express t Time Node j Reactive power regulation for critical nodes Sensitivity to voltage changes; and They are respectively t Time Node The active and reactive power regulation capabilities that can be provided; It consists of the sum of the node's active power up-adjustment potential and active power down-adjustment potential. It is given by the reactive power regulation potential of the node.

[0075] No. c Each cluster t Voltage regulation capability index at any time Defined as When the maximum voltage deviation within the cluster is less than the cluster's maximum voltage correction capability, the cluster can be considered to have sufficient voltage regulation capability. In this case, take... V,c,t =1. The system as a whole is in t Voltage regulation capability index at any time Take the average value of each cluster, that is:

[0076] In the formula, K t for t The total number of clusters corresponding to the current partitioning scheme at any given time. Clearly, V,t The larger the value, the stronger the ability of resources within each cluster to coordinate and adjust voltage deviations during the current period, and the more beneficial the partitioning result is to subsequent partition voltage coordination and control.

[0077] Step 2.7: Calculate the load fluctuation suppression capability index by combining the cluster net load fluctuation and active power regulation resource capacity. Specifically: To measure the cluster's ability to mitigate net load fluctuations, this invention introduces a load fluctuation evaluation window, using the change in cluster net load within the window as the evaluation object. c Each cluster during the time period t Net load fluctuation Represented as:

[0078] In the formula, and They are nodes i During the period t Active load and photovoltaic active power output;c Each cluster during the time period t The active power regulation capability that can be used to smooth out fluctuations is :

[0079] In the formula, Represents a node j During the period t The available active power regulation capacity is mainly provided by active power regulation resources such as energy storage.

[0080] No. c Each cluster within the window period t Load fluctuation suppression capability Represented as ,when When the time interval is 10, it indicates that there is no net load fluctuation in the cluster between adjacent time periods, and in this case, we take 10. When the active power regulation capacity within the cluster can cover the net load fluctuation, this indicator is also set to 1; otherwise, it is determined by the ratio of the two.

[0081] The system at any time t Overall load fluctuation suppression capability index Defined as In the formula, K t for t The total number of clusters corresponding to the current partitioning scheme at any given time; At the current moment t The load fluctuation evaluation window serves as the benchmark. This embodiment uses... This means that the cluster's ability to suppress volatility is measured by the change in net load over four consecutive time periods. Clearly, H,t The larger the value, the more conducive the current partitioning result is to leveraging the active power regulation resources within the cluster, thereby improving the ability to suppress net load fluctuations.

[0082] Step 2.8: Construct the time-varying comprehensive evaluation function, specifically as follows:

[0083] in, It is a time-varying comprehensive evaluation function. oh 1. oh 2 and oh 3 is the evaluation weight, which satisfies oh 1+ oh 2+ oh 3 = 1. It is a normalized modularity metric. Is the system as a whole in t Voltage regulation capability index at any given time. Is the system at any time t The overall load fluctuation suppression capability index.

[0084] A voltage risk factor is introduced to determine how close the node voltage is to the operating boundary during the current period:

[0085] In the formula, or V,t for t Voltage risk factor at any given moment; N This represents the set of nodes, excluding the balancing node, that participate in the voltage risk assessment. V max =1.05 pu is the upper limit of the node voltage. When or V,t When the value is less than 0.8, it indicates that the system voltage deviation is small. Cluster partitioning primarily emphasizes structural rationality and load fluctuation suppression capabilities. Therefore, a value of 0.8 is used.

[0086] when or V,t When the value is ≥0.8, it indicates that the voltage of some nodes is approaching the operating boundary, and the system has potential voltage risks. The weight of the voltage regulation capability index should be increased in cluster partitioning. At this point, [the value should be taken as...].

[0087] A voltage risk coefficient can be constructed based on how close the node voltage is to the operating boundary. When the voltage risk is low, the focus is on modularity and load fluctuation suppression capability; when the voltage risk is high, the focus is on voltage regulation capability.

[0088] Step 3: Using the candidate cluster partitioning schemes as the individual positions in the improved artificial hummingbird algorithm, and using the time-varying comprehensive evaluation function as the fitness criterion, the improved artificial hummingbird algorithm (AHA) is used to search for candidate cluster partitioning schemes in the current runtime; specifically: Step a: Initialize algorithm parameters; randomly generate a number of hummingbird individuals in the solution space as the initial population; Step b: The best individual from the previous time period, after being superimposed with random perturbation, is used as part of the initial population for the current time period; the formula is expressed as follows:

[0089] In the formula, It is the first in the initial population during the current time period. i Individual hummingbirds For boundary projection operators, The perturbation scale. It is a random perturbation vector.

[0090] Step c: Update the position of each individual hummingbird according to the following formula:

[0091] Where, η i For random mutation vectors, p m ( k ) is the first k The mutation probability is adjusted based on the number of iterations. K This represents the maximum number of iterations. It is the first k+ 1st generation i The location of an individual hummingbird It is the lower bound of the mutation probability. It is the upper limit of the mutation probability; Step d: After the location is updated, the candidate cluster partitioning scheme is discretized and its feasibility is corrected to meet the requirements of cluster number range, minimum cluster node number and network connectivity. Step e: Based on the comprehensive evaluation function c t Calculate the fitness value of each hummingbird, and select the one with the best fitness among all hummingbirds as the current optimal solution; Step f: Check if the termination condition is met. If not, directly retain one or more individuals with the best current fitness to the next generation and return to step c. If yes, output the current optimal solution as the cluster partitioning result.

[0092] Step 4: Evaluate and compare the new partitioning scheme obtained by the current runtime segment with the scheme that maintains the cluster structure of the previous time segment, and determine the final cluster partitioning result adopted by the current runtime segment based on the comparison result of the evaluation gain and the preset update threshold.

[0093] To avoid frequent changes in cluster boundaries due to minor operational disturbances, this invention simultaneously calculates the evaluation value of the newly obtained partitioning scheme in each time period. And the evaluation value of the cluster structure in the current state, based on the previous period. And define the evaluation gain:

[0094] Let the cluster structure update threshold be... s The cluster structure ultimately adopted in the current time period. satisfy: like ,but = ;otherwise = Through this dynamic update mechanism, the original cluster structure is maintained when the system operating status changes little, so as to improve the stability of partitioning; when the node voltage deviation increases, the net load fluctuation increases, or the resource margin changes significantly, the cluster boundary is updated again to improve the adaptability of the partitioning results to changes in operating status.

[0095] Step 5: Output the cluster partitioning results, which include node cluster labels, cluster boundary information, and branch cutting information, for active distribution network zonal optimization scheduling, intraday rolling optimization, or inter-cluster coordination control.

[0096] Example 2 This invention proposes an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the active distribution network dynamic cluster partitioning method based on the improved artificial hummingbird algorithm as described in Embodiment 1.

[0097] Example 3 This invention proposes a computer-readable storage medium storing a computer program that enables a computer to execute the active distribution network dynamic cluster partitioning method based on the improved artificial hummingbird algorithm as described in Embodiment 1.

[0098] An improved IEEE 33-node active distribution network was selected as the test system on the MATLAB simulation platform. The system integrates distributed photovoltaic (PV), SVC, CB, and ESS regulation resources. The allowable node voltage range is 0.95–1.05 pu, the scheduling cycle is 24 hours, and it is discretized into 96 operating segments in 15-minute intervals. Load for each time period is scaled using typical daily load curves, and distributed PV output is described using typical daily PV output curves. In each operating segment, the power flow results are updated based on the current load, PV output, and the operating status of regulation resources. Static electrical distance similarity, dynamic reactive power-voltage sensitivity similarity, and dynamic resource characteristic similarity are calculated respectively. In the comprehensive similarity matrix, the weights of the three similarity categories can be taken as follows: l e =0.2, l qv =0.4, l r =0.4. The comprehensive evaluation index of the cluster consists of modularity index, voltage regulation capability index, and load fluctuation suppression capability index. When the voltage risk coefficient is lower than the preset value, the evaluation weight can be [0.4, 0, 0.6]; when the voltage risk coefficient reaches or exceeds the preset value, the evaluation weight can be switched to [0.1, 0.7, 0.2]. The cluster structure update threshold can be 0.003, and the load fluctuation suppression capability is evaluated using a rolling window with a length of 4 time periods.

[0099] The improved artificial hummingbird algorithm can have a population size of 15, a maximum number of iterations of 20, a minimum cluster size of 3, a maximum cluster size of 6, and a minimum cluster node size of 2. The algorithm uses a comprehensive evaluation function as the fitness level, initializes the initial population quality using the optimal solution from the previous time step, preserves superior individuals using elite retention, and enhances search capabilities through adaptive mutation.

[0100] Figure 2 The dynamic partitioning results of active distribution network clusters obtained using the improved AHA algorithm are presented. As shown in the figure, the total number of system nodes remains constant throughout the 96 time periods of the day, but the node size of each cluster exhibits significant time-varying characteristics depending on the operating state. The proposed method can adaptively adjust the cluster boundaries according to changes in source load output and control requirements. Changes in dynamic cluster partitioning are mainly concentrated during periods of significant source load state transitions, such as 1:45–2:00, 5:45–7:15, and 16:00–18:00; while the cluster structure remains stable during relatively stable operating periods such as 2:00–5:30, 7:30–16:00, and 18:00–24:00. This demonstrates that the proposed improved AHA dynamic partitioning method can avoid frequent repartitioning while responding to changes in operating state, providing a stable and adaptive partitioning basis for subsequent intraday rolling optimization.

[0101] To verify the effectiveness of this invention, four schemes—dynamic partitioning, fixed partitioning, standard artificial hummingbird algorithm partitioning, and no partitioning—are compared. All four schemes use the same computational system, source-load time-series data, and evaluation metrics. The dynamic partitioning scheme is the proposed dynamic cluster partitioning method based on the improved AHA algorithm, which dynamically updates the cluster structure according to load, photovoltaic output, voltage sensitivity, and resource status at different times. The fixed partitioning scheme selects the partitioning results at typical moments as fixed partitions throughout the day; node affiliation does not change over time, but operational data is still updated according to the actual time sequence. The standard artificial hummingbird algorithm partitioning scheme uses the standard AHA algorithm for cluster partitioning without introducing improved mechanisms such as initialization of the previous time-series optimal solution, elite retention, and adaptive mutation, to verify the improved algorithm's effect on partitioning quality. The no-partitioning scheme evaluates the entire network as a whole, illustrating the necessity of cluster partitioning for active distribution network zonal control. The simulation results are shown in Table 1. The dynamic partitioning scheme can adjust the cluster boundary in a timely manner according to the changes in source load status and control requirements. Compared with fixed partitioning, standard artificial hummingbird algorithm partitioning, and no partitioning scheme, the dynamic cluster partitioning method of this invention is slightly lower than fixed partitioning and standard AHA partitioning in terms of modularity index, but achieves better results in terms of voltage regulation capability, fluctuation suppression capability, and comprehensive evaluation index. This shows that the method can better adapt to the time-varying operation scenario of active distribution network.

[0102] Table 1 Comparison of evaluation metrics under different cluster partitioning strategies

[0103] Figure 3 A comparison chart of voltage regulation capabilities under different cluster partitioning schemes. Figure 3 It can be seen that dynamic partitioning maintains high voltage regulation capability in multiple time periods, including the morning, noon, and evening, with its advantage becoming more pronounced, especially after dusk. In the fixed partitioning scheme, the node affiliation remains unchanged, but the load, photovoltaic output, node voltage, sensitivity, and resource margin still change over time, with the boundaries remaining constant throughout the day. This makes it difficult to adapt to changes in operating status caused by photovoltaic load decline and load increase, resulting in a significant decrease in voltage regulation capability in the later stages. While standard AHA partitioning performs well overall, it is still inferior to dynamic partitioning, indicating that combining improved AHA with a dynamic update mechanism can further enhance the cluster's adaptability to node voltage deviations. The no-partitioning scheme, lacking effective partition coordination, has the weakest overall voltage regulation performance.

[0104] Figure 4 A comparison chart of load fluctuation suppression capabilities under different cluster partitioning schemes. Figure 4 It can be seen that the no-division scheme showed high fluctuation suppression capability during some periods at noon, but its performance was poor at other times, with large overall fluctuations. The fixed division was generally low throughout the day, indicating that the fixed boundary could not continuously adapt to changes in net load fluctuation characteristics. Although the standard AHA division was better than the fixed division, it was still weaker than the dynamic division in the evening and night. In contrast, the dynamic division showed a more balanced fluctuation suppression capability throughout the day, indicating that it could adjust the cluster boundary in a timely manner according to changes in source load status, thereby better matching local net load characteristics.

[0105] In the embodiments disclosed in this application, a computer storage medium may be a tangible medium that may contain or store programs for use by or in conjunction with an instruction execution system, apparatus, or device. The computer storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of computer storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, and portable compact disc read-only memory (CD). ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0106] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0107] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A method for dynamic cluster partitioning of active distribution networks based on an improved artificial hummingbird algorithm, characterized in that, Includes the following steps: Based on active distribution network data, modularity index, voltage regulation capability index, and load fluctuation suppression capability index are established to construct a time-varying comprehensive evaluation function. The candidate cluster partitioning scheme is used as the individual position of the improved artificial hummingbird algorithm, and the time-varying comprehensive evaluation function is used as the fitness criterion. The improved artificial hummingbird algorithm is used to search for candidate cluster partitioning schemes in the current runtime. The new partitioning scheme obtained by the current runtime segment is evaluated and compared with the scheme that maintains the cluster structure of the previous time segment. The final cluster partitioning result adopted by the current runtime segment is determined based on the comparison result of the evaluation gain and the preset update threshold. The cluster partitioning result is output, which includes node cluster labels, cluster boundary information and branch cutting information, and is used for active distribution network zonal optimization scheduling, intraday rolling optimization or inter-cluster coordination control.

2. The method for dynamic cluster partitioning of active distribution networks based on the improved artificial hummingbird algorithm as described in claim 1, characterized in that, The establishment of modularity index, voltage regulation capability index, and load fluctuation suppression capability index based on active distribution network data specifically includes: Calculate the electrical path impedance between any two nodes based on topology data and line parameters, and construct a node static similarity matrix based on electrical distance; Based on the power flow calculation results of the current operating period, a power flow Jacobian matrix is ​​established, the approximate sensitivity relationship between reactive power injection changes and node voltage changes is extracted, and a node dynamic similarity matrix based on reactive power-voltage sensitivity is constructed. Based on the parameters of distributed photovoltaics, photovoltaic inverters, static var compensators, parallel capacitor banks, and the configuration and operational margin of energy storage devices at the nodes, a node resource feature vector is constructed, and a dynamic similarity matrix of resource features is constructed from the node resource feature vector. The node static similarity matrix, the node dynamic similarity matrix based on reactive power-voltage sensitivity, and the resource feature dynamic similarity matrix are weighted and fused to obtain the node comprehensive similarity matrix for the current operating segment. ; Based on the node comprehensive similarity matrix, a modularity index is established; The voltage regulation capability index is calculated by combining node voltage deviation, sensitivity relationship, and intra-cluster regulation resource margin. The load fluctuation suppression capability index is calculated by combining the cluster net load fluctuation and the active power regulation resource capacity.

3. The method for dynamic cluster partitioning of active distribution networks based on the improved artificial hummingbird algorithm as described in claim 2, characterized in that, The process of constructing the node static similarity matrix is ​​as follows: node i With nodes j The set of branches on the path between them is Ω. ij branch road ( m , n The impedance is Z mn Then the node i With nodes j Total impedance of the path between and electrical distance They are respectively: The electrical distance is mapped to node static similarity using a Gaussian kernel function: In the formula, For electrical distance scale parameters, It is a node i With nodes j Static similarity between nodes; traverse all nodes, and the static similarity between all nodes forms a node static similarity matrix. .

4. The method for dynamic cluster partitioning of active distribution networks based on the improved artificial hummingbird algorithm as described in claim 2, characterized in that, The process of constructing the node dynamic similarity matrix is ​​as follows: Perform power flow calculations during the current runtime to obtain the power flow Jacobian matrix. : In the formula, H This represents the sensitivity matrix of nodal active power injection to changes in voltage phase angle. N This represents the sensitivity matrix of nodal active power injection to changes in voltage amplitude. K This represents the sensitivity matrix of nodal reactive power injection to changes in voltage phase angle. L This represents the sensitivity matrix of nodal reactive power injection to changes in voltage amplitude; Based on the linearized power flow relationship, the node voltage increment U Changes in reactive power injection at nodes Q Connecting these points, we get: In the formula, This is an approximate reactive power-voltage sensitivity matrix. elements in Represents a node j Changes in reactive power injection on nodes i The degree of influence of voltage changes; Based on the approximate reactive power-voltage sensitivity matrix, the time interval is defined. t Next node i With nodes j Sensitivity difference index between for: In the formula, Represents a node i The degree to which changes in reactive power injection affect the voltage itself. Represents a node j The degree to which changes in reactive power injection affect the voltage itself. Represents a node j Reactive power injection changes on nodes i The degree of influence of voltage changes Represents a node i Reactive power injection changes on nodes j The degree of influence of voltage changes; Dynamic node similarity is constructed using a Gaussian kernel function: In the formula, σ qv This is a parameter for adjusting the reactive power-voltage sensitivity similarity. For nodes i With nodes j The dynamic similarity is calculated by traversing all nodes, and the dynamic similarity of all nodes forms a node dynamic similarity matrix. .

5. The method for dynamic cluster partitioning of active distribution networks based on the improved artificial hummingbird algorithm as described in claim 2, characterized in that, The specific steps for constructing node resource feature vectors and then building a dynamic similarity matrix of resource features from these node resource feature vectors are as follows: Define nodes i At any moment t resource feature vector for: In the formula, Represents a node i Contributing to photovoltaic power generation; Indicates the photovoltaic inverter at time t Maximum adjustable reactive power capacity; Represents a node i The rated apparent capacity of the photovoltaic inverter; Indicates that the photovoltaic inverter is at the node i time t Those who have made meritorious contributions; Represents a node i The reactive power adjustable quantity of the SVC; Represents a node i The maximum reactive power compensation capacity of the parallel capacitor bank, of which The maximum number of groups that can be added. For single-group compensation capacity; and Representing nodes respectively i The maximum charging and discharging power and rated capacity of the energy storage device are specified. If the node is not configured with the corresponding type of resource, the corresponding component is set to 0. After normalizing the features of each dimension, it is defined at time... t Next node i With nodes j Differences in resource characteristics between for: In the formula, and Representing nodes respectively i With nodes j At any moment t Normalized resource feature vector; Represents the L2 norm; time t Next node i With nodes j Resource feature similarity between for: In the formula, As the resource feature similarity adjustment parameter, we iterate through all nodes, and the resource feature similarity of all nodes forms a dynamic resource feature similarity matrix. .

6. The method for dynamic cluster partitioning of active distribution networks based on the improved artificial hummingbird algorithm as described in claim 2, characterized in that, The modularity index established based on the node comprehensive similarity matrix is ​​specifically as follows: Time t The following is a comprehensive similarity matrix of nodes Viewed as a weighted undirected network, it is called a comprehensive similarity network, where the network nodes are distribution network nodes participating in cluster partitioning, and the edge weights are... Represents a node i With nodes j The overall similarity between them defines the nodes. i Weighting degree in the network for: The total edge weight of the network is then expressed as: In the formula, N This represents the total number of nodes participating in the cluster partitioning; express t At any given moment i The weighting degree; m t Indicates in t The total edge weight in a comprehensive similar network at any given time; Set time t Next node i Cluster number c i,t If node i With nodes j If they are in the same cluster, then the cluster similarity / difference identifier is used. δ ( c i,t , c j,t )=1, otherwise δ ( c i,t , c j,t )=0; the original modularity corresponding to the partitioning scheme. Represented as: In the formula, express t At any given moment j The weighted degree is used to normalize the original modularity index to obtain the normalized modularity index. .

7. The method for dynamic cluster partitioning of active distribution networks based on the improved artificial hummingbird algorithm as described in claim 2, characterized in that, The specific voltage regulation capability index is as follows: set up t Time of the first c The set of nodes in each cluster is Ω c,t The maximum voltage deviation within this cluster is expressed as: , will the c In each cluster t The node with the largest absolute value of voltage deviation at any given time is defined as the critical node. The maximum voltage correction capability of resources within the cluster for this critical node. Represented as: In the formula, express t Time Node j Active power regulation for key nodes Approximate sensitivity to voltage changes express t Time Node j Reactive power regulation for critical nodes Sensitivity to voltage changes; and They are respectively t Time Node The active and reactive power regulation capabilities that can be provided; No. c Each cluster t Voltage regulation capability index at any time Defined as The system as a whole is t Voltage regulation capability index at any time Take the average value of each cluster; The specific index for calculating load fluctuation suppression capability is as follows: No. c Each cluster during the time period τ Net load fluctuation Represented as: In the formula, and They are nodes i During the period τ Active load and photovoltaic active power output; No. c Each cluster during the time period τ The active power regulation capability that can be used to smooth out fluctuations is , No. c Each cluster within the window period τ Load fluctuation suppression capability Represented as The system is at all times t Overall load fluctuation suppression capability index Defined as In the formula, K t for t The total number of clusters corresponding to the current partitioning scheme at any given time; At the current moment t A benchmark load fluctuation evaluation window.

8. The method for dynamic cluster partitioning of active distribution networks based on the improved artificial hummingbird algorithm as described in claim 1, characterized in that, The time-varying comprehensive evaluation function is specifically as follows: in, It is a time-varying comprehensive evaluation function. ω 1. ω 2 and ω 3 represents the evaluation weight. It is a normalized modularity metric. Is the system as a whole in t Voltage regulation capability index at any given time. Is the system at any time t The overall load fluctuation suppression capability index; The process of using the improved artificial hummingbird algorithm to search for candidate cluster partitioning schemes in the current runtime segment specifically involves: Step a: Initialize algorithm parameters; randomly generate a number of hummingbird individuals in the solution space as the initial population; Step b: The best individual from the previous time period is superimposed with random perturbation and used as part of the initial population for the current time period; Step c: Update the position of each individual hummingbird according to the following formula: Where, η i For random mutation vectors, p m ( k ) is the first k The mutation probability is adjusted based on the number of iterations. K This represents the maximum number of iterations. It is the first k+ 1st generation i The location of an individual hummingbird It is the lower bound of the mutation probability. It is the upper limit of the mutation probability; For boundary projection operators; Step d: After the location is updated, the candidate cluster partitioning scheme is discretized and its feasibility is corrected to meet the requirements of cluster number range, minimum cluster node number and network connectivity. Step e: Based on the comprehensive evaluation function γ t Calculate the fitness value of each hummingbird, and select the one with the best fitness among all hummingbirds as the current optimal solution; Step f: Check if the termination condition is met. If not, directly retain one or more individuals with the best current fitness to the next generation and return to step c. If yes, output the current optimal solution as the cluster partitioning result.

9. An electronic device, characterized in that, include: The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the active distribution network dynamic cluster partitioning method based on the improved artificial hummingbird algorithm as described in any one of claims 1-8.

10. A computer-readable storage medium storing a computer program, characterized in that, The computer program causes the computer to execute the active distribution network dynamic cluster partitioning method based on the improved artificial hummingbird algorithm as described in any one of claims 1-8.

Images