A flexible power distribution network fault recovery method based on topology feature fusion

Abstract

The application discloses a flexible power distribution network fault recovery method based on topology feature fusion, and particularly relates to the field of power system fault recovery. When a flexible interconnected power distribution network fails, island division of a fault side power distribution network, power distribution network reconstruction and adjustment of control modes of each SOP are comprehensively performed to recover power supply. The aggregated topology features of the fault side power distribution network are calculated to assist a genetic algorithm in solving, wherein; for the case that the power distribution network simultaneously contains SOPs and distributed power sources, corresponding island power flow calculation and island division methods are proposed; and the application is helpful to better tap the conditional capacity of the flexible interconnected power distribution network by adjusting the influence of the SOP control mode on the recovery strategy.

Description

Technical Field

[0001] This application relates to the field of power system fault recovery, and more specifically, to a method for flexible distribution network fault recovery based on topology feature fusion. Background Technology

[0002] With the gradual popularization and integration of distributed generation and flexible interconnection devices such as soft switching, the power flow regulation and load transfer capabilities of distribution networks have been significantly improved. Furthermore, they can supply power to loads independently in an islanded state without connection to the upstream grid. Based on this, the concept of active distribution networks has been proposed. How to better allocate restoration resources within the distribution system and optimize the power restoration results is currently the main research focus. Distribution network fault recovery is a multi-objective, multi-constraint nonlinear optimization problem. The final solution is a series of switch state combinations. This set of switch states can redistribute system power when a system fault occurs, achieving the function of restoring power supply.

[0003] Introducing flexible interconnect devices requires additional consideration of the operational characteristics of flexible switches. However, introducing new variables can lead to problems such as increased search space and premature convergence of the population. To address this, this invention uses a graph Laplacian operator to aggregate node features during the recovery process and uses the aggregated vector to obtain the next generation of individuals, thereby improving the algorithm's performance.

[0004] Therefore, a flexible distribution network fault recovery method based on topology feature fusion is proposed to address the above problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, this application provides a flexible distribution network fault recovery method based on topology feature fusion to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, this application provides the following technical solution: a flexible distribution network fault recovery method based on topological feature fusion. When a fault occurs in a flexible interconnected distribution network, the method assists a genetic algorithm in solving the problem by calculating the aggregated topological features of the fault-side distribution network.

[0007] Under the premise of SOP multi-control mode, a power supply restoration model is established with the weighted value of node voltage drop, total power loss load and number of switching actions as the objective function and as the index for evaluating the restoration effect.

[0008] For distribution networks that contain both SOPs (Standard Operating Programs) and distributed generation, a corresponding islanding power flow calculation and islanding method is proposed.

[0009] Preferably, after a distribution network fault occurs, the load nodes downstream of the fault are in a state of power loss and need to be re-energized. The objective function minimizes the total load loss in the distribution network, expressed as:

[0010] minf1=∑ i∈N w i c i p i (t),

[0011] Where f1 is the total load loss, N is the total number of nodes in the distribution network, and w i Let c be the load level of node i. i P is a 0-1 variable representing whether node i is powerless. i (t) represents the active power of node i during time period t;

[0012] While ensuring load restoration, it is also necessary to ensure that the restored topology and power flow distribution are reasonable. This requires minimizing system line losses and node voltage deviations to achieve the goal of optimizing power flow distribution, specifically expressed as follows:

[0013] minf2=∑ i∈B p loss,i ,

[0014] Where f2 represents the main bus loss of the distribution network, B represents the total number of branches in the distribution network, and P loss,i This represents the line loss on the i-th branch;

[0015] In a hybrid AC / DC distribution network, both the AC and DC networks are required to have radial constraints. To ensure that the restored topology is a radial network, an additional penalty, f3, is needed when a closed loop is detected in the network. Specifically:

[0016]

[0017] In addition, to reduce the switching costs of fault recovery and to tap the islanding potential of SOPs and distributed power sources, a corresponding penalty amount f4 is set for whether the line switch status changes.

[0018]

[0019] Where D is the cost of changing the switching state of a line, x(i) represents the on / off state of the i-th line after recovery, and B f Let k(i) represent the faulty line, and k(i) represent the on / off state of the i-th line before the fault.

[0020] The final objective function is obtained by weighting f1-f4, with weights of μ1-μ4.

[0021] minf = u1f1 + u2f2 + u3f3 + u4f4.

[0022] Preferably, the constraints of the power restoration model include:

[0023] a. The power flow constraints of AC and DC networks are as follows:

[0024]

[0025] Among them, Q i The reactive power of node i during time period (t)t, G ij With B ij θ represents the conductance and susceptance between node i and node j. ij (t) represents the phase angle difference between node i and node j in time period t;

[0026] According to Kirchhoff's current law and the principle of active power conservation, the power flow constraints of a DC network are expressed as follows:

[0027]

[0028] Where S is the number of AC and DC interconnection ports, L is the number of DC network nodes, and P is the number of DC network nodes. si P dk These represent the power exchanged between the AC system and the DC system at the i-th port, and the DC load or power source at the k-th port, respectively, with the injection into the DC network being the positive direction.

[0029] b. After the fault is recovered, the voltage of each node is within a reasonable range, which is set between 0.95 and 1.02 times the rated voltage; the current flow on each branch is less than the maximum allowable power.

[0030] 0.95U n ≤U i (t)≤1.02U n ,

[0031] P l (t)≤P lmax

[0032] Among them, P l (t) represents the active power flowing through the l-th line at time t, P lmax This represents the maximum active power carried by the line.

[0033] c. Based on the back-to-back VSC type flexible soft switch, pulse width modulation technology is used for control. As can be seen from the principle of pulse width modulation, by adjusting the phase angle and modulation degree of the modulation wave, the voltage and hysteresis angle of the AC side of the VSC are changed, thereby flexibly controlling the active and reactive components of the flexible soft switch and realizing four-quadrant operation in the PQ plane.

[0034] VSC's control modes are as follows:

[0035] Master-slave control method: Maintain P ei Q ei Constant, P ei Q ei These represent the active power and reactive power injected into the AC system by the SOP, respectively.

[0036] Droop control method: Maintain P ei (t) is constant, Q ei The relationship between (t) and the voltage amplitude of the SOP access node is as follows:

[0037] Q ei =(1.02U) n -U ei (t)) / λ i ,

[0038] λ i =0.07U n / Q max ,

[0039] Among them, U ei (t) represents the voltage amplitude of the i-th SOP connected to the AC system node during time period t, and represents the droop coefficient of the i-th node;

[0040] Maintain U di (t), Q ei (t) is constant, where U di (t) represents the DC voltage of the i-th SOP during time period t;

[0041] v / f control mode: Maintain U ei (t), θ ei (t) is constant, where θ ei (t) represents the voltage phase angle of the i-th SOP connected to the AC system node during time period t;

[0042] d. Islanding Constraint: In an islanded operating region, the sum of the active power of the distributed power sources and the power transferred from the SOP should be greater than the sum of the total load and bus losses in the region.

[0043]

[0044] Where I represents the total number of isolated islands in the communication section, and n j Let P be the number of distributed power sources in the j-th island. i DG,j (t) represents the output of the i-th distributed power source during time period t, where n s,j Let P be the number of SOPs in the j-th isolated island. ei,i(t) represents the power supplied by the i-th SOP to the island, N j Let P be the total number of nodes in the j-th isolated island. i P(t) represents the active power of the i-th node at time t. j loss (t) represents the bus loss of the j-th island at time t.

[0045] Preferably, in order to maintain the above constraints, corresponding islanding and load shedding operations are required. The specific steps are as follows:

[0046] Step 1: Obtain initial data for the distribution network: network topology, impedance, load, distributed generation output, and load priority recovery factor;

[0047] Step 2: Based on the network topology, obtain the number of islands and the island nodes they contain. If there is no distributed power source or SOP in an island area, all nodes in the island area are included as power-out nodes. If there are more than two SOPs or adjustable distributed power sources in an island area, select the node with the higher active power output limit as the balancing node of this island.

[0048] Step 3: Include all load nodes within the island and perform power flow calculations;

[0049] Step 4: Verify whether the power flow calculation results meet the constraints;

[0050] Step 5: If the condition is met, proceed to Step 7; otherwise, proceed to Step 6.

[0051] Step 6: Sort all load nodes by load level, cut off the lowest level and smallest capacity load, and return to Step 4;

[0052] Step 7: Complete the division of the isolated islands.

[0053] Preferably, for areas that have not lost power, power flow calculations need to be performed by including all load nodes within the island, and the results of the power flow calculations need to be verified to ensure that the node voltage and line power flow meet the relevant constraints.

[0054] Preferably, the basic idea of ​​topology feature aggregation is that when the feature amplitude of a node is lower than that of other nodes in its neighborhood, the feature amplitude of the central node should be further reduced; when it is higher than that of other nodes in its neighborhood, the feature amplitude of the central node should be further increased, thereby amplifying the feature differences between different nodes; thus realizing the characterization of the topology of the distribution network through voltage and power.

[0055] The specific steps are as follows:

[0056] a. Construct a topological adjacency matrix Aij A ij Represented as:

[0057]

[0058] Construct a degree matrix, which is a diagonal matrix whose diagonal elements are the reciprocals of the square roots of the corresponding rows in the adjacency matrix, represented as:

[0059]

[0060] The calculation method for the Graph Laplace operator is as follows:

[0061]

[0062] b. Select node characteristics as follows: node voltage amplitude, from the parent node; if the node is located in an island, then it is the node closer to the power supply point, and the active and reactive power of the injected branches form an n*3 node characteristic matrix X. ij Applying the graph Laplacian operator to X ij Then, column normalization is performed to obtain the aggregated node feature matrix X. * ij ,Right now:

[0063]

[0064] Where, m 3×3 (L ij X ij ) indicates L ij X ij A 3×3 diagonal matrix composed of the maximum values ​​of each column.

[0065] Preferably, the specific steps of improving the genetic algorithm based on the aggregated topological features are as follows: During the crossover process, for two individuals C to be exchanged... a With C b Calculate the 2-norm difference between each row vector in the aggregated node feature matrix to form an n-dimensional aggregated feature difference vector:

[0066]

[0067] Where, x a,1i x a,2i x a,3i ∈C a x b,1i x b,2i x b,3i ∈C b After that, according to D iBased on the calculation results, a roulette wheel algorithm is used to select the exchange node. The chromosome position represented by the upstream branch of the node is used as the exchange position, and the chromosome exchange operation is performed. That is, the larger the topological feature difference of the node after aggregation, the higher the probability that the chromosome position represented by its upstream branch will be selected as the crossover point.

[0068] The technical effects and advantages of this application are as follows:

[0069] 1. Compared with existing technologies, this flexible distribution network fault recovery method based on topology feature fusion proposes a model solution method combining topology aggregation features for the fault recovery problem of flexible distribution networks containing VSC-type SOPs. This method uses back-to-back VSC-type SOPs as interconnecting devices and considers four control modes, including master-slave and droop control, as optimization variables. To address the problems of increased search space caused by introducing SOP control variables, premature convergence of metaheuristic algorithms, and slow convergence, a topology feature aggregation method based on non-Euclidean graph convolution is adopted. Node voltage, active power, and reactive power are selected as node features, and the difference in node feature vectors after aggregation of the previous generation is used as the basis for generating the next generation.

[0070] Compared with existing technologies, this flexible distribution network fault recovery method based on topological feature fusion improves the aggregation topological feature difference between the new individuals after crossover and the previous generation individuals by adopting a crossover strategy guided by topological feature aggregation in the genetic algorithm solution process, compared with the crossover method based solely on random numbers. The aggregation topological feature difference includes differences in topological structure and power flow distribution. Searching with the goal of generating greater differences can better avoid the problem of premature convergence of the population in a suboptimal recovery result.

[0071] Compared with existing technologies, this flexible distribution network fault recovery method based on topology feature fusion helps to better explore the conditional capabilities of flexible interconnected distribution networks by adjusting the SOP control mode to influence the recovery strategy.

[0072] This method uses back-to-back VSC-type SOPs as interconnection devices and takes into account four control modes, including master-slave and drooping, as optimization variables. To address the problems of increased search space caused by the introduction of SOP control variables, premature convergence and slow convergence of metaheuristic algorithms, a topological feature aggregation method based on non-Euclidean graph convolution is adopted. Node voltage, active power and reactive power are selected as node features, and the difference in node feature vectors after aggregation of the previous generation individuals is used as the basis for generating the offspring individuals. Attached Figure Description

[0073] Figure 1 This is a schematic diagram of the fault recovery method of this application;

[0074] Figure 2 This is a diagram of the dual IEEE 33 flexible interconnect system of this application;

[0075] Figure 3 This diagram illustrates the efficiency of the algorithm using topological feature aggregation and not using topological feature set in this application. Detailed Implementation

[0076] 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 skilled in the art without creative effort are within the scope of protection of this application.

[0077] Example:

[0078] As attached Figures 1 to 3 This paper presents a fault recovery method for flexible distribution networks based on topology feature fusion.

[0079] To better illustrate this fault recovery method, the following examples are provided: Figure 2 As shown, based on the IEEE 33-node distribution network system, two IEEE 33-node systems are flexibly interconnected through two dual-port SOPs. The load of distribution network 1 is set to be 1.1 times that of distribution network 2. The DC network contains one DC distributed power source and one DC load with sizes of 150kW and 50kW, respectively. The SOP access nodes are nodes 14 and 32 in the two distribution networks. The upper limit of the output power of the SOP at node 14 is 250kW, and the upper limit of the output power of the SOP at node 7 is 100kW. The backup tie switches in distribution network 1 are set to be 8-21, 9-15, and 33-25. The line 5-6 is set to be disconnected due to a fault.

[0080] A power restoration model is constructed, using a weighted average of node voltage drop, total power loss load, and number of switch actions as the objective function, which serves as the indicator for evaluating the restoration effect. Specifically:

[0081] minf1=∑ i∈N w i c i p i (t),

[0082] Where f1 is the total load loss. wi is the load level of node i, and c i Let Pi(t) be a 0-1 variable representing whether node i is de-energized. Pi(t) is the active power of node i during time period t.

[0083] minf2=∑ i∈B p loss,i ,

[0084] Where f2 represents the bus loss of the distribution network, P loss,i This represents the line loss on the i-th branch.

[0085]

[0086] In addition, to reduce the switching costs of fault recovery and to tap the islanding potential of SOPs and distributed power sources, a corresponding penalty amount f4 is set for whether the line switch status changes.

[0087]

[0088] Where D is the cost of changing the switching state of a line, x(i) represents the on / off state of the i-th line after recovery, and B f The faulty line is represented by k(i), which represents the on / off state of the i-th line before the fault.

[0089] The final objective function is obtained by weighting f1-f4, with weights ranging from μ1 to μ4.

[0090] minf = u1f1 + u2f2 + u3f3 + u4f4.

[0091] Step 1: Construct a power restoration model, considering constraints such as node voltage and islanding, specifically:

[0092] a. Power flow constraints in AC and DC networks:

[0093]

[0094] Among them, Q i The reactive power of node i during time period (t)t, G ij With B ij θ represents the conductance and susceptance between node i and node j. ij (t) represents the phase angle difference between node i and node j in time period t;

[0095] According to Kirchhoff's current law and the principle of active power conservation, the power flow constraints of a DC network are expressed as follows:

[0096]

[0097] Where S is the number of AC and DC interconnection ports, L is the number of DC network nodes, and P is the number of DC network nodes. si P dk These represent the power exchanged between the AC system and the DC system at the i-th port, and the DC load or power source at the k-th port, respectively, with the injection into the DC network being the positive direction.

[0098] b. After the fault is recovered, the voltage of each node is within a reasonable range, which is set between 0.95 and 1.02 times the rated voltage; the current flow on each branch is less than the maximum allowable power.

[0099] 0.95U n ≤U i (t)≤1.02U n ,

[0100] P l (t)≤P lmax

[0101] Among them, P l (t) represents the active power flowing through the l-th line at time t, P lmax This represents the maximum active power carried by the line.

[0102] c. Based on the back-to-back VSC type flexible soft switch, pulse width modulation technology is used for control. As can be seen from the principle of pulse width modulation, by adjusting the phase angle and modulation degree of the modulation wave, the voltage and hysteresis angle of the AC side of the VSC are changed, thereby flexibly controlling the active and reactive components of the flexible soft switch and realizing four-quadrant operation in the PQ plane.

[0103] VSC's control modes are as follows:

[0104] Master-slave control method: Maintain P ei Q ei Constant, P ei Q ei These represent the active power and reactive power injected into the AC system by the SOP, respectively.

[0105] Droop control method: Maintain P ei (t) is constant, Q ei The relationship between (t) and the voltage amplitude of the SOP access node is as follows:

[0106] Q ei =(1.02U) n -U ei (t)) / λ i ,

[0107] λ i =0.07U n / Q max ,

[0108] Among them, U ei (t) represents the voltage amplitude of the i-th SOP connected to the AC system node during time period t, and represents the droop coefficient of the i-th node;

[0109] Maintain U di (t), Q ei (t) is constant, where Udi (t) represents the DC voltage of the i-th SOP during time period t;

[0110] v / f control mode: Maintain U ei (t), θ ei (t) is constant, where θ ei (t) represents the voltage phase angle of the i-th SOP connected to the AC system node during time period t;

[0111] d. Islanding Constraint: In an islanded operating region, the sum of the active power of the distributed power sources and the power transferred from the SOP should be greater than the sum of the total load and bus losses in the region.

[0112]

[0113] Where I represents the total number of isolated islands in the communication section, and n j Let P be the number of distributed power sources in the j-th island. i DG,j (t) represents the output of the i-th distributed power source during time period t, where n s,j Let P be the number of SOPs in the j-th isolated island. ei,i (t) represents the power supplied by the i-th SOP to the island, N j Let P be the total number of nodes in the j-th isolated island. i P(t) represents the active power of the i-th node at time t. j loss (t) represents the bus loss of the j-th island at time t.

[0114] To maintain the above constraints, corresponding islanding and load shedding operations are required. The specific steps are as follows:

[0115] Step 1: Obtain initial data for the distribution network: network topology, impedance, load, distributed generation output, and load priority recovery factor;

[0116] Step 2: Based on the network topology, obtain the number of islands and the island nodes they contain. If there is no distributed power source or SOP in an island area, all nodes in the island area are included as power-out nodes. If there are more than two SOPs or adjustable distributed power sources in an island area, select the node with the higher active power output limit as the balancing node of this island.

[0117] Step 3: Include all load nodes within the island and perform power flow calculations;

[0118] Step 4: Verify whether the power flow calculation results meet the constraints;

[0119] Step 5: If the condition is met, proceed to Step 7; otherwise, proceed to Step 6.

[0120] Step 6: Sort all load nodes by load level, cut off the lowest level and smallest capacity load, and return to Step 4;

[0121] Step 7: Complete the division of the isolated islands.

[0122] For areas that have not lost power, it is necessary to perform power flow calculations by including all load nodes within the island, and verify whether the results of the power flow calculations meet the constraints to ensure that the node voltage and line power flow meet the relevant constraints.

[0123] The basic idea of ​​topology feature aggregation is that when the feature amplitude of a node is lower than that of other nodes in its neighborhood, the feature amplitude of this central node should be further reduced; when it is higher than that of other nodes in its neighborhood, the feature amplitude of the central node should be further increased, thereby amplifying the feature differences between different nodes; thus, the topology of the distribution network can be characterized by voltage and power.

[0124] The specific steps are as follows:

[0125] a. Construct a topological adjacency matrix A ij A ij Represented as:

[0126]

[0127] Construct a degree matrix, which is a diagonal matrix whose diagonal elements are the reciprocals of the square roots of the corresponding rows in the adjacency matrix, represented as:

[0128]

[0129] The calculation method for the Graph Laplace operator is as follows:

[0130]

[0131] b. Select node characteristics as follows: node voltage amplitude, from the parent node; if the node is located in an island, then it is the node closer to the power supply point, and the active and reactive power of the injected branches form an n*3 node characteristic matrix X. ij Applying the graph Laplacian operator to X ij Then, column normalization is performed to obtain the aggregated node feature matrix X. * ij ,Right now:

[0132]

[0133] Where, m 3×3 (L ij X ij ) indicates L ij Xij A 3×3 diagonal matrix composed of the maximum values ​​of each column.

[0134] The specific steps for improving the genetic algorithm based on the topological characteristics after aggregation are as follows: During the crossover process, for two individuals C to be exchanged... a With C b Calculate the 2-norm difference between each row vector in the aggregated node feature matrix to form an n-dimensional aggregated feature difference vector:

[0135]

[0136] Where, x a,1i x a,2i x a,3i ∈C a x b,1i x b,2i x b,3i ∈C b After that, according to D i Based on the calculation results, a roulette wheel algorithm is used to select the exchange node. The chromosome position represented by the upstream branch of the node is used as the exchange position, and the chromosome exchange operation is performed. That is, the larger the topological feature difference of the node after aggregation, the higher the probability that the chromosome position represented by its upstream branch will be selected as the crossover point.

[0137] The results of the standard genetic algorithm and the results of the algorithm proposed in this application are as follows: Figure 3 As shown, although the algorithm proposed in this paper converges slightly later than the standard genetic algorithm, it still achieves the best convergence result when the initial population is poor.

[0138] The implementation principle of this application embodiment is as follows: First, when a fault occurs in the flexible interconnected distribution network, the islanding of the fault-side distribution network, the reconfiguration of the distribution network, and the adjustment of the control mode of each SOP are integrated to restore power supply. The aggregated topology characteristics of the fault-side distribution network are calculated to assist the genetic algorithm in solving the problem. Under the premise of multiple control modes of SOP, a power supply restoration model considering power flow balance constraints, node voltage constraints, and branch load constraints is adopted. The weighted values ​​of node voltage drop, total power loss load, and number of switching actions are used as the objective function and as the index for evaluating the restoration effect. For the case where the distribution network contains both SOP and distributed generation, a corresponding islanding power flow calculation and islanding method is proposed.

Claims

1. A fault recovery method for flexible distribution networks based on topology feature fusion, characterized in that: When a fault occurs in a flexible interconnected distribution network, the aggregated topology characteristics of the fault-side distribution network are calculated to assist the genetic algorithm in solving the problem. First, establish a power restoration model; Specifically, under the premise of SOP multi-control mode, a power supply restoration model is established with the weighted value of node voltage drop, total power loss load and number of switching actions as the objective function and as the indicator for evaluating the restoration effect. For distribution networks that simultaneously contain both SOPs and distributed generation, corresponding islanding power flow calculation and islanding methods are proposed. For areas that have not lost power, it is necessary to perform power flow calculations by including all load nodes within the island and verify whether the results of the power flow calculations meet the constraints to ensure that the node voltage and line power flow meet the relevant constraints. Topological feature aggregation involves further reducing the feature amplitude of a central node when its feature amplitude is lower than that of other nodes in its neighborhood, and further increasing its feature amplitude when its feature amplitude is higher than that of other nodes in its neighborhood, thereby amplifying the feature differences between different nodes. This allows for the characterization of the distribution network topology through voltage and power. The specific steps are as follows: a. Construct a topological adjacency matrix A ij A ij Represented as: Construct a degree matrix, which is a diagonal matrix whose diagonal elements are the reciprocals of the square roots of the corresponding rows in the adjacency matrix, represented as: The calculation method for the Graph Laplace operator is as follows: b. Select the node characteristics as: node voltage amplitude, active power and reactive power injected into the node's branches from the superior node, forming an n*3 node characteristic matrix X. ij Applying the graph Laplacian operator to X ij Then, column normalization is performed to obtain the aggregated node feature matrix X. * ij ,Right now: Where, m 3×3 (L ij X ij ) indicates L ij X ij A 3×3 diagonal matrix composed of the maximum values ​​of each column.

2. The method for fault recovery of flexible distribution networks based on topology feature fusion according to claim 1, characterized in that: After a distribution network fault occurs, the load nodes downstream of the fault are in a state of power loss and need to be re-energized. The objective function minimizes the total load loss in the distribution network, expressed as: minf1=Σ i∈N w i c i p i (t), Where f1 is the total load loss, N is the total number of nodes in the distribution network, and w i Let c be the load level of node i. i P is a 0-1 variable representing whether node i is powerless. i (t) represents the active power of node i during time period t; While ensuring load restoration, it is also necessary to ensure that the restored topology and power flow distribution are reasonable. This requires minimizing system line losses and node voltage deviations to achieve the goal of optimizing power flow distribution, specifically expressed as follows: minf2=Σ i∈B p loss,i , Where f2 represents the main bus loss of the distribution network, B represents the total number of branches in the distribution network, and P loss,i This represents the line loss on the i-th branch; In a hybrid AC / DC distribution network, both the AC and DC networks are required to have radial constraints. To ensure that the restored topology is a radial network, an additional penalty, f3, is needed when a closed loop is detected in the network. Specifically: In addition, to reduce the switching costs of fault recovery and to tap the islanding potential of SOPs and distributed power sources, a corresponding penalty amount f4 is set for whether the line switch status changes. Where D is the cost of changing the switching state of a line, x(i) represents the on / off state of the i-th line after recovery, and B f Let k(i) represent the faulty line, and k(i) represent the on / off state of the i-th line before the fault. The final objective function is obtained by weighting f1-f4, with weights of μ1-μ4. minf = u1f1 + u2f2 + u3f3 + u4f4.

3. The method for fault recovery of flexible distribution networks based on topology feature fusion according to claim 2, characterized in that: The constraints of the power restoration model include: a. The power flow constraints of AC and DC networks are as follows: Among them, Q i The reactive power of node i during time period (t)t, G ij With B ij θ represents the conductance and susceptance between node i and node j. ij (t) represents the phase angle difference between node i and node j in time period t; According to Kirchhoff's current law and the principle of active power conservation, the power flow constraints of a DC network are expressed as follows: Where S is the number of AC and DC interconnection ports, L is the number of DC network nodes, and P is the number of DC network nodes. si P dk These represent the power exchanged between the AC system and the DC system at the i-th port, and the DC load or power source at the k-th port, respectively, with the injection into the DC network being the positive direction. b. After the fault is recovered, the voltage of each node is within a reasonable range, which is set between 0.95 and 1.02 times the rated voltage; the current flow on each branch is less than the maximum allowable power. 0.95U n ≤U i (t)≤1.02U n , P l (t)≤Pl max Among them, P l (t) represents the active power flowing through the l-th line at time t, P lmax This represents the maximum active power carried by the line. c. Based on the back-to-back VSC type flexible soft switch, pulse width modulation technology is used for control. As can be seen from the principle of pulse width modulation, by adjusting the phase angle and modulation degree of the modulation wave, the voltage and hysteresis angle of the AC side of the VSC are changed, thereby flexibly controlling the active and reactive components of the flexible soft switch and realizing four-quadrant operation in the PQ plane. VSC's control modes are as follows: Master-slave control method: Maintain P ei Q ei Constant, P ei Q ei These represent the active power and reactive power injected into the AC system by the SOP, respectively. Droop control method: Maintain P ei (t) is constant, Q ei The relationship between (t) and the voltage amplitude of the SOP access node is as follows: Q ei =(1.02U n -U ei (t)) / λ i , l i =0.07U n / Q max , Among them, U ei (t) represents the voltage amplitude of the i-th SOP connected to the AC system node during time period t, and represents the droop coefficient of the i-th node; Maintain U di (t), Q ei (t) is constant, where U di (t) represents the DC voltage of the i-th SOP during time period t; v / f control mode: Maintain U ei (t), θ ei (t) is constant, where θ ei (t) represents the voltage phase angle of the i-th SOP connected to the AC system node during time period t; d. Islanding Constraint: In an islanded operating region, the sum of the active power of the distributed power sources and the power transferred from the SOP should be greater than the sum of the total load and bus losses in the region. Where I represents the total number of isolated islands in the communication section, and n j Let J be the number of distributed power sources in the j-th isolated island. Let n be the output of the i-th distributed power source during time period t. s,j Let P be the number of SOPs in the j-th isolated island. ei,i (t) represents the power supplied by the i-th SOP to the island, N j Let P be the total number of nodes in the j-th isolated island. i P(t) represents the active power of the i-th node at time t. j loss (t) represents the bus loss of the j-th island at time t.

4. The method for fault recovery of flexible distribution networks based on topology feature fusion according to claim 3, characterized in that: To maintain the above constraints, corresponding islanding and load shedding operations are required. The specific steps are as follows: Step 1: Obtain initial data for the distribution network: network topology, impedance, load, distributed generation output, and load priority recovery factor; Step 2: Based on the network topology, obtain the number of islands and the island nodes they contain. If there is no distributed power source or SOP in an island area, all nodes in the island area are included as power-out nodes. If there are more than two SOPs or adjustable distributed power sources in an island area, select the node with the higher active power output limit as the balancing node of this island. Step 3: Include all load nodes within the island and perform power flow calculations; Step 4: Verify whether the power flow calculation results meet the constraints; Step 5: If the condition is met, proceed to Step 7; otherwise, proceed to Step 6. Step 6: Sort all load nodes by load level, cut off the lowest level and smallest capacity load, and return to Step 4; Step 7: Complete the division of the isolated islands.

5. The method for fault recovery of flexible distribution networks based on topology feature fusion according to claim 1, characterized in that: The specific steps for improving the genetic algorithm based on the topological characteristics after aggregation are as follows: During the crossover process, for two individuals C to be exchanged... a With C b Calculate the 2-norm difference between each row vector in the aggregated node feature matrix to form an n-dimensional aggregated feature difference vector: Where, x a,1i x a,2i x a,3i ∈C a x b,1i x b,2i x b,3i ∈C b After that, according to D i The calculation results are used to select the exchange node using the roulette wheel algorithm. The chromosome position represented by the upstream branch of the node is used as the exchange position to perform the chromosome exchange operation. That is, the larger the topological feature difference of the node after aggregation, the higher the probability that the chromosome position represented by its upstream branch will be selected as the crossover point.

Images