Two-stage divided active power distribution network reconstruction method

Abstract

The invention relates to the technical field of power systems, in particular to a two-stage division active power distribution network reconstruction method, which comprises the following steps of: proposing an objective function F based on power loss; a two-stage ADNR scheme is adopted; performing network division based on hierarchical spectral clustering; and performing two-stage optimization. By combining power loss minimization, a two-stage reconstruction scheme, hierarchical spectral clustering and an optimization technology, an efficient power distribution network reconstruction strategy is provided, the power loss can be reduced, and the operation efficiency of the power distribution network can be improved; the method can adapt to the dynamic change of the power distribution network under the background that distributed power generation is increasingly popularized, and improves the stability and reliability of the system.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a two-level divided active distribution network reconstruction method. Background Art

[0002] With the widespread application of distributed generators (DG) in power systems, traditional distribution networks are facing new challenges, especially in terms of how to improve the operating efficiency and stability of distribution systems. In order to better adapt to the access of distributed generation, active distribution network (ADN) reconstruction has become an effective means to optimize the topology of distribution networks, reduce power losses, and improve system stability and power supply reliability. Network reconstruction reconfigures the topology of distribution networks by reasonably opening or closing switching devices in the distribution network, thereby achieving optimal load distribution and efficient transmission of electric energy.

[0003] At present, the distribution network reconstruction methods are mainly divided into three categories: heuristic methods, mathematical programming (MP) methods and meta-heuristic-based algorithms. Heuristic methods usually reconstruct the network through strategies such as branch exchange. Although they are simple and easy to use, they take a long time to process and have low optimization accuracy. Mathematical programming-based methods rely on the precise expressions of the objective function and constraints. They usually solve mixed integer nonlinear programming (MINLP) problems, but the solution process is complex and the computational cost is high. Meta-heuristic algorithms seek global optimal solutions through random search. Although they have good global search capabilities, they still need to be improved in terms of solution accuracy and computational efficiency.

[0004] Although existing methods can optimize the operation of distribution networks to a certain extent, how to quickly and effectively reconfigure the network in the isolated island operation mode during the integration of distributed generation is still an urgent problem to be solved. Especially in emergency situations, how to form self-sufficient islands in the shortest time and protect other areas of the network from interference has become a key challenge to achieve rapid response and efficient operation of distribution networks. Summary of the invention

[0005] In view of at least one of the above technical problems, the present invention provides a two-level divided active distribution network reconstruction method, which adopts the improvement of the method to improve the operation efficiency of the distribution network.

[0006] According to a first aspect of the present invention, a two-level divided active power distribution network reconstruction method is provided, comprising the following steps:

[0007] Propose an objective function F based on power loss;

[0008] A two-stage ADNR program was adopted;

[0009] Network partitioning based on hierarchical spectral clustering;

[0010] Perform two-stage optimization.

[0011] In some embodiments of the present invention, the objective function F is as follows:

[0012]

[0013] In the formula, h varies from 1 to 24; is the weighting factor of the line loss of branch ij; ij is the switch status between branches ij, 1 means closed, 0 means disconnected; K is the weighting factor, K is set to 1 in normal operation and to 0 in fault conditions; E i is the power status of bus i, 1 means power on, 0 means no power on; is the voltage amplitude at bus i; is the voltage amplitude at bus j; is the voltage angle at busbar i; is the voltage angle at busbar j; is the actual power flow through branch j at hour h; R ij and X ij are the resistance and reactance of branch ij respectively; w LM is the weighting factor assigned to the load size; β i is the priority of bus i for load; w LP is the weighting factor assigned to the load priority; is the active power demand in hour h; s b is the set of branches; n b Is a collection of buses.

[0014] In some embodiments of the present invention, the objective function is limited by the balance of active power and reactive power in the network, and the active power expression is:

[0015]

[0016] The reactive power expression is:

[0017]

[0018] In the formula, is the active power generation of bus i; is the demand of bus i; is the reactive power generation of bus i; is the demand of busbar i; G ij is the real part of the i-th and j-th columns of the admittance matrix; B ij is the imaginary part of the i-th and j-th columns of the admittance matrix; N B Is a collection of buses.

[0019] In some embodiments of the present invention, the ADNR scheme includes node active power generation and node reactive power generation, and the node active power generation is expressed as:

[0020]

[0021] In the formula, is the active power produced by the conventional generator at bus i; is the active power generated by the distributed generator on bus i;

[0022] The reactive power generation of the node is:

[0023]

[0024] In the formula, is the reactive power produced by the conventional generator at bus i; is the reactive power generated by the distributed generator on bus i.

[0025] In some embodiments of the present invention, the ADNR scheme also includes limiting the current of each distribution line. Voltage Frequency f h and distributed generator capacity, the constraint formula is as follows:

[0026]

[0027] The superscripts min and max represent the upper and lower limits of the corresponding variables respectively; f h is the distribution line frequency; and They are the rated capacity, maximum active power and maximum reactive power of distributed generators respectively.

[0028] In some embodiments of the present invention, in order to satisfy the radiality and connectivity of the network, the following formula is used:

[0029]

[0030] Where N is the number of buses; Ns is the number of closed switches including sectionalizer S and tie switch TW; ss ) represents any busbar i and substation busbar i ss The distance between.

[0031] In some embodiments of the present invention, when performing network partitioning based on hierarchical spectral clustering, given information about buses and branches, an undirected graph related to distribution network reconstruction can be constructed as G(v x ,e), the bus set is represented by v x, the branch set is represented as e, and then the entire network is divided and reconfigured using hierarchical spectral clustering. The reconfiguration includes the following steps:

[0032] Perform dynamic edge weighting and calculate edge weights based on topology and admittance;

[0033] Compute the normalized Laplacian operator;

[0034] Select the first k eigenvectors of the eigenvalues ​​for spectral embedding;

[0035] Compute hierarchical clustering based on the Euclidean distance of feature vectors.

[0036] In some embodiments of the present invention, the topology calculation is used to measure the pure connectivity of the network, that is, ij =1 means branch ij is connected, Where (i, j)∈e represents the branch from bus i to j;

[0037] The admittance calculation is used to measure the strength of the network connection, i.e., w ij =Y ij , (i, j)∈e, where Y ij is the admittance of the edge from vertex i to j;

[0038] According to the actual power flow G(v x ,e) Each edge is assigned a weight, the formula is:

[0039]

[0040] In the formula, w ij =(|P ij |+|P ji |) / 2;P ij is the actual power flowing through branch ij from bus i, P ji is the actual power flowing through branch ji from bus j; N is the number of buses;

[0041] The calculation of the normalized Laplacian operator L N Use the following formula:

[0042]

[0043] In the formula, w ii (w jj ) is G(v x ,e) the dynamic allocation weight of edge ii(jj);

[0044] Calculate L N The formula for the characteristic value λ is: 0 = λ 1 ≤λ 2 ≤...≤λ Nand its corresponding eigenvector Ψ 1 ,Ψ 2 ,...,Ψ N , spectral clustering using L N The first k eigenvectors of The N buses in the dimension provide the coordinates, and the value of k is chosen in the range of 2 to N based on the highest value of the relative intrinsic gap EG, as follows:

[0045]

[0046] Where k means that a good network decomposition allows at least k partitions; EG k is the relative feature gap when the number of clusters is k; k (λ k-1 ) is the k(k-1)th eigenvalue;

[0047] When calculating hierarchical clustering based on the Euclidean distance of feature vectors, the solution vector is normalized to R k The length of is 1, and the formula is:

[0048]

[0049] In the formula, x i is the i-th row of X, which is an N×k matrix whose columns contain the first k eigenvectors corresponding to the eigenvalues; u i is the solution vector x i Normalized value.

[0050] In some embodiments of the present invention, when performing two-stage optimization, it is divided into a first stage and a second stage. The first stage is to solve the objective function F to obtain a unique switching state for each sub-network, that is, min where r m is the internal edge set in the mth SN, and the switching state of each subnetwork can be written as γ m =[γ ij ; (i,j)∈r m ];

[0051] In the second stage, the objective function F is solved again to obtain the switching state of the boundary edge, that is, Where b is the set of boundary edges. In the second stage, the solution obtained from the first stage is Keep it unchanged and solve for S 2 =γ, where γ = [γ ij ; (i, j)∈b], where S 2 The second stage is to solve the state of the boundary edge switches of each sub-network.

[0052] In some embodiments of the present invention, using an improved particle swarm algorithm to solve the optimization problems of the first stage and the second stage includes the following steps:

[0053] Expand the search space;

[0054] Speed ​​up your search.

[0055] The beneficial effects of the present invention are as follows: the present invention provides an efficient distribution network reconstruction strategy by combining power loss minimization, a two-stage reconstruction scheme, hierarchical spectral clustering and optimization technology, which can not only reduce power loss and improve the operating efficiency of the distribution network; compared with the existing technology, it can adapt to the dynamic changes of the distribution network and improve the stability and reliability of the system under the background of the increasing popularity of distributed power generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0057] Figure 1 It is a flowchart of the steps of the two-level divided active distribution network reconstruction method in an embodiment of the present invention;

[0058] Figure 2 A flowchart of the steps of reconfiguring the entire network partition using hierarchical spectral clustering in an embodiment of the present invention;

[0059] Figure 3 The present invention is a flowchart of the steps of using the particle swarm algorithm to solve the optimization problems of the first and second stages. DETAILED DESCRIPTION

[0060] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0061] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0062] like Figures 1 to 3 The two-level divided active distribution network reconstruction method shown includes the following steps:

[0063] S10: Propose an objective function F based on power loss;

[0064] S20: adopt a two-stage ADNR scheme;

[0065] S30: Network partitioning based on hierarchical spectral clustering;

[0066] S40: Perform two-stage optimization.

[0067] like Figure 1 As shown, in step S10, power loss is one of the key factors in the optimization design of the distribution network. By defining power loss as the objective function, the energy loss in the system can be minimized during the network reconstruction process, thereby improving the overall efficiency of the power system. This objective function F usually takes into account line loss, transformer loss and other factors that may affect the operating efficiency of the system.

[0068] In step S20, it should be noted that ADNR, namely active distribution network reconstruction, is a distribution network reconstruction method that takes dynamic demand and load changes into consideration.

[0069] In step S30, hierarchical spectral clustering is a commonly used clustering method, which can divide the distribution network into multiple sub-areas based on the similarity between network nodes (such as the similarity of current, voltage or load). This division method usually helps to further optimize the overall structure of the distribution network. In active distribution networks, hierarchical spectral clustering can effectively identify key areas, perform centralized or distributed optimization, and improve the stability and efficiency of the system.

[0070] In step S40, based on the network division, two-stage optimization is performed, which can be divided into firstly performing local optimization and then performing global optimization.

[0071] In the above embodiments, the present invention provides an efficient distribution network reconstruction strategy by combining power loss minimization, a two-stage reconstruction scheme, hierarchical spectral clustering and optimization technology, which can not only reduce power loss and improve the operating efficiency of the distribution network; compared with the existing technology, it can adapt to the dynamic changes of the distribution network and improve the stability and reliability of the system under the background of the increasing popularity of distributed power generation.

[0072] In an embodiment of the present invention, the objective function F is as follows:

[0073]

[0074] In the formula, h varies from 1 to 24; is the weighting factor of the line loss of branch ij; ijis the switch status between branches ij, 1 means closed, 0 means disconnected; K is the weighting factor, K is set to 1 in normal operation and to 0 in fault conditions; E i is the power status of bus i, 1 means power on, 0 means no power on; is the voltage amplitude at bus i; is the voltage amplitude at bus j; is the voltage angle at busbar i; is the voltage angle at busbar j; is the actual power flow through branch j at hour h; R ij and X ij are the resistance and reactance of branch ij respectively; w LM is the weighting factor assigned to the load size; β i is the priority of bus i for load; w LP is the weighting factor assigned to the load priority; is the active power demand in hour h; s b is the set of branches; n b is the set of buses. In normal operation, K is set to 1, so F is connected by using w ij The biased line segments take into account line efficiency, not just the associated line losses. However, in the case of a fault, setting K to 0 results in an islanded mode of operation. The priorities of the loads in the island formed by the network partition are then checked after the fault occurs. If the island contains any priority loads, then w LM and w LP Set to 0 and 1 respectively.

[0075] The objective function F comprehensively considers various factors in the distribution network, including power loss, switch status, voltage control, etc. By weighted summation of these factors, the optimization algorithm can be adjusted according to the power flow and loss conditions of different time periods and branches to achieve optimal reconstruction of the distribution network, reduce power loss and improve the operating efficiency of the system.

[0076] In an embodiment of the present invention, in order to ensure the power supply and demand balance in the network and the stable operation of the system, the objective function is limited to the balance of active power and reactive power in the network, and the active power expression is:

[0077]

[0078] The reactive power expression is:

[0079]

[0080] In the formula, is the active power generation of bus i; is the demand of bus i; is the reactive power generation of bus i; is the demand of busbar i; G ij is the real part of the i-th and j-th columns of the admittance matrix; B ij is the imaginary part of the i-th and j-th columns of the admittance matrix; N B is a collection of buses. Active power and balance constraints are the core constraints for distribution network reconstruction, ensuring that the power supply and demand of all nodes in the system are balanced and the network can operate normally after optimization and adjustment. Through the above two constraints, the optimization goal of the objective function can be achieved while satisfying the physical constraints.

[0081] In an embodiment of the present invention, the ADNR scheme includes node active power generation and node reactive power generation, and the node active power generation is expressed as:

[0082]

[0083] In the formula, is the active power produced by the conventional generator at bus i; is the active power generated by the distributed generator on bus i;

[0084] The node reactive power generation is:

[0085]

[0086] In the formula, is the reactive power produced by the conventional generator at bus i; It is the reactive power generated by the distributed generator on bus i. Distributed generation equipment can generate electricity directly near the load, thereby reducing losses during power transmission. It can also adjust the power generation according to real-time demand and cooperate with traditional generators to optimize the power distribution of the entire distribution network. In the ADNR scheme, traditional power generation and distributed power generation need to work together. Traditional power generation provides large-scale and stable active power, and at the same time provides reactive power by adjusting the excitation system to maintain voltage stability. The active power generation and reactive power generation of the nodes in the ANDR scheme are provided by traditional generators and distributed generators respectively. This combination of power generation methods can effectively optimize the power distribution of the distribution network, reduce energy loss, and improve the reliability and flexibility of system operation, meeting the multi-objective optimization needs of modern active distribution networks.

[0087] In an embodiment of the present invention, the ADNR scheme also includes limiting the current of each distribution line. Voltage Frequency f h and distributed generator capacity, the constraint formula is as follows:

[0088]

[0089] The superscripts min and max represent the upper and lower limits of the corresponding variables respectively; f h is the distribution line frequency; and They are the rated capacity, maximum active power and maximum reactive power of distributed generators. Under the above constraints, the constraints of current, voltage and frequency prevent overload, overvoltage or excessive frequency fluctuations in the operation of the distribution network; the capacity limit of distributed generators protects the equipment from overload or overload operation, etc. The above constraint formula provides the physical boundary conditions for safe and stable operation of the ADNR scheme. Through strict restrictions on current, voltage, frequency and capacity of distributed generators, the distribution network can meet the load demand while avoiding equipment overload or system instability, providing a basic guarantee for the realization of the optimization goal.

[0090] On the basis of the above embodiment, in order to meet the radiality and connectivity of the network, the following formula is adopted:

[0091]

[0092] Where N is the number of buses; Ns is the number of closed switches including sectionalizer S and tie switch TW; ss ) represents any busbar i and substation busbar i ss The distribution network needs to meet radiality, that is, the topology of the entire network is similar to a tree network, that is, a loop-free connected network, with only one unique path between every two nodes. The radial setting makes it easy to locate faults, and any line faults can be quickly isolated without affecting the power supply of other parts, and can simplify control. The loop-free structure reduces the complexity of power distribution and reduces possible power oscillations. In addition, the role of connectivity constraints is to ensure the integrity of network operation, that is, all loads (busbars) can obtain power supply support from substations and avoid local power outages or island operation problems caused by the lack of connection of some nodes. The sectionalizer and tie switch mentioned in the above constraint formula are key components for achieving radiality and connectivity. The sectionalizer is used to isolate certain parts of the network and cut off the power flow when a fault occurs, which facilitates fault isolation. The tie switch is used to establish a new power flow path to meet connectivity during normal operation or reconstruction. The closed state of these switches directly affects the topology of the distribution network and determines whether the network meets the radiality and connectivity requirements. These two constraints jointly ensure the structural rationality, power supply reliability and operational safety of the distribution network.

[0093] In an embodiment of the present invention, when performing network partitioning based on hierarchical spectral clustering, given information about buses and branches, an undirected graph related to distribution network reconstruction can be constructed as G(v x ,e), the bus set is represented by vx , the branch set is denoted as e, and then the entire network is partitioned and reconfigured using hierarchical spectral clustering. Due to the simplicity and availability of spectral clustering in practical systems, it is also applicable to various other power system-oriented applications, such as controlled islanding and emergency classification, e.g. Figure 2 As shown, reconfiguration includes the following steps:

[0094] S31: Perform dynamic edge weighting and calculate the edge weight based on topology and admittance;

[0095] S32: Calculate the normalized Laplacian operator;

[0096] S33: Select the first k eigenvectors of the eigenvalues ​​for spectral embedding;

[0097] S34: Calculate hierarchical clustering based on the Euclidean distance of feature vectors.

[0098] In step S31, in the topological structure, the edge weight can reflect the physical connection between the busbars, such as the distance, the connected equipment, etc.; in the admittance calculation, the admittance value can reflect the electrical characteristics of the line and affect the power transmission efficiency.

[0099] In step S32, the Laplacian operator can capture the structural characteristics of the graph, and its eigenvalues ​​and eigenvectors play an important role in network partitioning.

[0100] In step S33, spectral embedding is to map the nodes in the graph to a low-dimensional space through the eigenvalues ​​and eigenvectors of the Laplace operator, the purpose of which is to retain the structural characteristics of the image in the low-dimensional space and lay the foundation for subsequent clustering.

[0101] In step S34, hierarchical clustering is performed based on the Euclidean distance of the feature vectors in the low-dimensional space. The distance between each two nodes is determined by the Euclidean distance between their embedded feature vectors.

[0102] On the basis of the above embodiments, the method based on hierarchical spectral clustering completes the efficient division of the distribution network through dynamic edge weight calculation, Laplace operator, spectral embedding and hierarchical clustering. This method combines the physical characteristics of the network with mathematical tools, provides reliable theoretical support for the reconstruction of the distribution network, and helps to achieve the optimal configuration and efficient operation of the distribution system.

[0103] In the embodiment of the present invention, topology calculation is used to measure the pure connectivity of the network, that is, ij =1 means branch ij is connected, Where (i, j)∈e represents the branch from bus i to j; topological calculation is mainly used to construct the initial structural information of the network and provide a connectivity basis for subsequent weight allocation and spectral clustering.

[0104] Admittance calculation is used to measure the strength of the network connection, that is, w ij =Y ij , (i, j)∈e, where Y ij is the admittance of the edge from vertex i to j; the weight calculated by admittance can reflect the physical characteristics and impedance characteristics of power transmission, and is an important basis for optimizing network division. The above two formulas are static weights that remain unchanged in the network. Therefore, dynamic weighting is applied to G(v according to the actual power flow in the following way x ,e) Assign a weight to each edge.

[0105] According to the actual power flow G(v x ,e) Each edge is assigned a weight, the formula is:

[0106]

[0107] In the formula, w ij =(|P ij |+|P ji |) / 2;P ij is the actual power flowing through branch ij from bus i, P ji is the actual power flowing through branch ji from bus j; N is the number of buses; this weight calculation method takes into account the dynamic power flow characteristics, making the network division closer to the actual operating state. This formula is a dynamic weight that changes relative to the actual operating conditions of the network, so the dynamics of distributed generators are also taken into account. w ij Smaller values ​​indicate less power flow through the line.

[0108] Calculate the normalized Laplacian L N Use the following formula:

[0109]

[0110] In the formula, w ii (w jj ) is G(v x ,e)’s edge ii(jj)’s dynamic weight assignment; the normalized Laplace operator can better balance the structural characteristics of the network and is especially suitable for processing networks with dynamic weights.

[0111] Calculate L N The formula for the characteristic value λ is: 0 = λ 1 ≤λ 2 ≤...≤λ N and its corresponding eigenvector Ψ 1 ,Ψ 2 ,...,Ψ N , spectral clustering using L N The first k eigenvectors of The N buses in the dimension provide the coordinates, and the value of k is chosen in the range of 2 to N based on the highest value of the relative intrinsic gap EG, as follows:

[0112]

[0113] Where k means that a good network decomposition allows at least k partitions; EG k is the relative feature gap when the number of clusters is k; k (λ k-1 ) is the k(k-1)th eigenvalue; the maximum value of the relative eigengap ensures the partition quality of the network and makes the partition result have clear physical meaning.

[0114] When calculating hierarchical clustering based on the Euclidean distance of feature vectors, the solution vector is normalized to R k The length of is 1, and the formula is:

[0115]

[0116] In the formula, x i is the i-th row of X, which is an N×k matrix whose columns contain the first k eigenvectors corresponding to the eigenvalues; u i is the solution vector x i Normalized value; Normalization makes the coordinates of each node in low-dimensional space unified on the unit sphere, which helps to improve the accuracy of hierarchical clustering. The above description details how to calculate dynamic weights based on topology, admittance and power flow characteristics, select the optimal number of clusters through normalized Laplace operator and eigenvalue analysis, and finally use hierarchical spectral clustering to optimize the division of the distribution network. This method combines physical meaning with mathematical tools and is suitable for complex distribution network reconstruction scenarios.

[0117] In the embodiment of the present invention, when performing two-stage optimization, it is divided into a first stage and a second stage. The first stage is to solve the objective function F to obtain a unique switching state for each sub-network, that is, min where r m is the internal edge set in the mth SN, and the switching state of each subnetwork can be written as γ m =[γ ij ; (i,j)∈r m ]. When performing the first stage optimization, the following two rules need to be followed: 1. If the boundary edge of each subnetwork includes a tie switch, the tie switch state is assumed to be closed; 2. At least one bus in each subnetwork should be connected to the distributed generator connection bus or the virtual slack bus.

[0118] In the second stage, the objective function F is solved again to obtain the switching state of the boundary edge, that is, Where b is the set of boundary edges. In the second stage, the solution obtained from the first stage Keep it unchanged and solve for S 2 =γ, where γ = [γ ij ; (i, j)∈b], where S 2 is the state of the boundary edge switches of each subnetwork in the second stage. The entire optimization process is completed in two stages, and the exact solution of load flow and switch is obtained after the second stage. Therefore, the total computation time of the proposed method is the sum of the time required in the first stage (obtaining γ 1 arrive The maximum computation time required) and the second stage (the computation time required to obtain γ).

[0119] The first stage solves the problem of internal switching states, and the structure of each subnetwork is determined in this stage; the second stage focuses on the connection problem between subnetworks, and keeps the solution of the first stage fixed in this stage, with the goal of optimizing the connection edges between subnetworks. This method divides the problem into two parts, internal and boundary, and gradually optimizes it, so that the complex global optimization problem is transformed into multiple local optimization problems, simplifying the calculation and solution process. In this embodiment, by gradually adjusting the internal structure and boundary structure of the subnetwork, the entire network can be optimized, and the reliability and operation efficiency of the system can be improved.

[0120] In the embodiment of the present invention, an improved particle swarm algorithm is used to solve the optimization problems of the first stage and the second stage. The traditional particle swarm algorithm is prone to fall into a local optimal solution, especially when the search space is complex and has a high dimension. Figure 3 As shown in FIG. 1 , the improved particle swarm algorithm introduces the mutation operator of the genetic algorithm and the strategy of accelerating the search into the traditional algorithm, thereby improving the diversity and speed of the search, including the following steps:

[0121] S41: Expand the search space;

[0122] S42: Accelerate the search.

[0123] In step S41, the original particle swarm algorithm falls into a local optimum due to the knowledge sharing mechanism with other particles. Therefore, in order to reduce the possibility of premature convergence, the mutation operator of the genetic algorithm is used to modify the original particle swarm algorithm. The mutation point is from the dimension R W ×C L A random selection from the group matrix, where R W represents the size of the group, C L Represents a group that is positioned to represent the number of switch positions. This is the idea of ​​modifying the group position with a uniformly generated random number within the allowed range of the dimension, which increases the flexibility of the original particle swarm algorithm to search outside the main area of ​​the search space.

[0124] In step S42, in order to speed up the search, each swarm particle moves toward the current optimal position by calculating the average of two random swarms. The first convergence criterion (CC 1 ) based on the maximum change in the global fitness value being less than the specified tolerance of 10 -6 . The first convergence criterion is set. When the maximum change of the global fitness value is less than the specified tolerance, the algorithm is considered to have converged. The fitness value represents the quality of the particle swarm search solution, and the maximum change is a measure of the convergence degree of the search process. If the fitness change is too small, it means that the movement of the particles tends to be stable, the search process has been basically completed, and the iteration can be stopped. In this embodiment, the improved particle swarm algorithm effectively solves the problem that the traditional particle swarm algorithm is prone to fall into local optimality and slow convergence in high-dimensional complex optimization problems by expanding the search space and accelerating the search strategy. Expanding the search space uses the mutation operator of the genetic algorithm to increase diversity in the initial stage; accelerated search speeds up convergence by calculating the average position of two particles. Combining these two methods, the improved particle swarm algorithm can better balance exploration and development when solving optimization problems such as distribution network reconstruction, and achieve more efficient optimization effects.

[0125] Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A two-level partitioned active distribution network reconstruction method, characterized in that: The following steps are involved: Propose an objective function F based on power loss; A two-stage ADNR program was adopted; Network partitioning based on hierarchical spectral clustering; Perform two-stage optimization.

2. The two-level divided active distribution network reconstruction method according to claim 1 is characterized in that: The objective function F is as follows: In the formula, h varies from 1 to 24; is the weighting factor of the line loss of branch ij; ij is the switch status between branches ij, 1 means closed, 0 means disconnected; K is the weighting factor, K is set to 1 in normal operation and to 0 in fault conditions; E i is the power-on state at bus i, 1 means power on, 0 means no power on; V i h is the voltage amplitude at bus i; is the voltage amplitude at bus j; is the voltage angle at busbar i; is the voltage angle at busbar j; is the actual power flow through branch j at hour h; R ij and X ij are the resistance and reactance of branch ij respectively; w LM is the weighting factor assigned to the load size; β i is the priority of bus i for load; w LP is the weighting factor assigned to the load priority; P i D,h is the active power demand in hour h; s b is the set of branches; n b Is a collection of buses.

3. The two-level divided active distribution network reconstruction method according to claim 2 is characterized in that: The objective function is limited by the active power and reactive power balance in the network, and the active power expression is: The reactive power expression is: In the formula, P i G,h is the active power generation of bus i; P i D,h is the demand of bus i; is the reactive power generation of bus i; is the demand of busbar i; G ij is the real part of the i-th and j-th columns of the admittance matrix; B ij is the imaginary part of the i-th and j-th columns of the admittance matrix; N B Is a collection of buses.

4. The two-level divided active distribution network reconstruction method according to claim 3 is characterized in that: The ADNR scheme includes node active power generation and node reactive power generation, and the node active power generation is expressed as: P i G,h =P i C,h +P i DG,h ; Where P i C,h is the active power generated by the traditional generator at bus i; P i DG,h is the active power generated by the distributed generator on bus i; The reactive power generation of the node is: In the formula, Q i C,h is the reactive power produced by the conventional generator at bus i; is the reactive power generated by the distributed generator on bus i.

5. The two-level divided active distribution network reconstruction method according to claim 4 is characterized in that: The ADNR scheme also includes limiting the current of each distribution line. Voltage V i h , frequency f h and distributed generator capacity, the constraint formula is as follows: f min ≤f h ≤f max ; The superscripts min and max represent the upper and lower limits of the corresponding variables respectively; f h is the distribution line frequency; and They are the rated capacity, maximum active power and maximum reactive power of distributed generators respectively.

6. The two-level divided active distribution network reconstruction method according to claim 5 is characterized in that: In order to meet the radiality and connectivity of the network, the following formula is adopted: Ns = N-1; Where N is the number of buses; Ns is the number of closed switches including sectionalizer S and tie switch TW; ss ) represents any busbar i and substation busbar i ss The distance between.

7. The two-level divided active distribution network reconstruction method according to claim 1 is characterized in that: When performing network partitioning based on hierarchical spectral clustering, given the information about the bus and the branch, the undirected graph related to the distribution network reconstruction can be constructed as G(v x ,e), the bus set is represented by v x , the branch set is represented as e, and then the entire network is divided and reconfigured using hierarchical spectral clustering. The reconfiguration includes the following steps: Perform dynamic edge weighting and calculate edge weights based on topology and admittance; Compute the normalized Laplacian operator; Select the first k eigenvectors of the eigenvalues ​​for spectral embedding; Compute hierarchical clustering based on the Euclidean distance of feature vectors.

8. The two-level divided active distribution network reconstruction method according to claim 7 is characterized in that: The topological calculation is used to measure the pure connectivity of the network, that is, w ij =1 means branch ij is connected, Where (i, j)∈e represents the branch from bus i to j; The admittance calculation is used to measure the strength of the network connection, i.e., w ij =Y ij , (i, j)∈e, where Y ij is the admittance of the edge from vertex i to j; According to the actual power flow G(v x ,e) Each edge is assigned a weight, the formula is: In the formula, wij=(Pij|+Pji) / 2; P ij is the actual power flowing through branch ij from bus i, P ji is the actual power flowing through branch ji from bus j; N is the number of buses; The calculation of the normalized Laplacian operator L N Use the following formula: In the formula, w ii (w jj ) is G(v x ,e) the dynamic allocation weight of edge ii(jj); Calculate L N The formula for the characteristic value λ is: 0=λ1≤λ2≤...≤λ N and its corresponding eigenvectors Ψ1, Ψ2, ..., Ψ N , spectral clustering using L N The first k eigenvectors of The N buses in the dimension provide the coordinates, and the value of k is chosen in the range of 2 to N based on the highest value of the relative intrinsic gap EG, as follows: Where k means that a good network decomposition allows at least k partitions; EG k is the relative feature gap when the number of clusters is k; λ k (λ k-1 ) is the k(k-1)th eigenvalue; When calculating hierarchical clustering based on the Euclidean distance of feature vectors, the solution vector is normalized to R k The length of is 1, and the formula is: In the formula, x i is the i-th row of X, which is an N-by-k matrix whose columns contain the first k eigenvectors corresponding to the eigenvalues; u i is the solution vector x i Normalized value.

9. The two-level divided active distribution network reconstruction method according to claim 6, characterized in that: When performing two-stage optimization, it is divided into the first stage and the second stage. The first stage is to solve the objective function F to obtain the unique switching state of each sub-network, that is, where r m is the internal edge set in the mth SN, and the switching state of each subnetwork can be written as γ m =[γ ij ; (i,j)∈r m ]; In the second stage, the objective function F is solved again to obtain the switching state of the boundary edge, that is, Where b is the set of boundary edges. In the second stage, the solution obtained from the first stage is Keep it unchanged and solve for S 2 =γ, where γ = [γ ij ; (i, j)∈b], where S 2 The second stage is to solve the state of the boundary edge switches of each sub-network.

10. The two-level divided active distribution network reconstruction method according to claim 9, characterized in that: The improved particle swarm algorithm is used to solve the optimization problems of the first stage and the second stage, including the following steps: Expand the search space; Speed ​​up your search.

Images