A distribution network reconstruction method and system
Through Bayesian statistical inference distribution estimation calculation method, the switching state of the distribution network is optimized, and the problems of grid loss and voltage constraints in distribution network reconstruction are solved, and efficient distribution network reconstruction and voltage quality improvement are achieved.
Patent Information
- Application Number
- CN201910706429.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2019-08-01
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2039-08-01
AI Technical Summary
When optimizing the network structure, the existing distribution network reconstruction method is difficult to effectively reduce the network loss, while meeting the current, voltage and radiation constraints, and has low calculation efficiency.
Using the distribution estimation algorithm (IEDA) based on Bayesian statistical inference, the switching state of the distribution network is optimized by establishing a priori, conditional and posterior probability models, and the switching state of the distribution network is optimized, using roulette method and sampling technology to generate new individuals, and the group is updated until the constraints are met. , realize the reconstruction of the distribution network.
It significantly reduces the active loss of the distribution network, improves the voltage quality, optimizes the network structure, has a fast calculation speed, strong adaptability, and meets multiple constraints.
Smart Images

Figure CN110474324B_ABST
Abstract
Description
Technical Field
[0001] The present application belongs to the technical field of distribution network, and specifically relates to a distribution network reconstruction method and system. Background Art
[0002] Distribution network reconstruction is to optimize the network structure and reduce network losses under the premise of ensuring that the distribution network is radial and that the voltage and power flow do not exceed the limit. The essence of the reconstruction problem is a combinatorial optimization problem with multiple constraints. At present, the methods used for distribution network reconstruction mainly include expert system methods, mathematical programming methods, heuristic methods, genetic algorithms, etc. These methods have their own advantages and disadvantages. The distribution estimation algorithm (EDA) is also called a genetic algorithm based on a probability model. It uses the "global control" operation mode to replace the genetic operators in the genetic algorithm that have a destructive effect on the "building blocks". Because the distribution estimation algorithm has a stronger theoretical basis, it has been valued by many scholars and has become a research hotspot in the current field of evolutionary computing. Bayesian statistical inference theory is an important branch of mathematical statistics. Unlike the Bayesian network optimization method, the Bayesian statistical inference method does not need to optimize a complex network structure, but directly uses the information provided by the sample to infer the sample by establishing a posterior probability distribution. Therefore, it can speed up the EDA algorithm. Summary of the invention
[0003] The present application provides a distribution network reconstruction method, comprising:
[0004] ① Establish the main program module and initialize the population: first close all the switch branches involved in the coding in the distribution network, and then randomly open a branch switch involved in the coding in each loop in turn until the total number of opened branches is equal to the total number of initial connecting switches in the network. Through this branch opening operation, an individual is generated, and then stored in the initial population after passing the inspection by the network topology analysis module in the main program. When the number of individuals in the initial population reaches the value m, stop generating the initial population; the value m is related to the specific scale of the distribution network. The larger the network scale, the larger the required value m, and the smaller the network scale, the smaller the value m. For a distribution network with n nodes and k connecting branches, the constraint condition is 0<m<C(n-1,k)-3C(n-2,k-1), where C(n-1,k) is the number of combinations;
[0005] ② Update the population, establish the IEDA module, and set the population update rules: first determine the number of new individuals that the algorithm requires to be updated, and then sample the posterior probability model. The sampling method is based on each column of the updated posterior probability, and the dimension number with a larger probability of taking zero value for the corresponding dimension variable of the new individual is naturally selected until the number of zero-value variables selected is equal to the number of distribution network loops. The individuals are repeatedly sampled using the roulette method, and the new individuals and the current dominant population are combined into the next generation population. The formula is used Calculate the fitness of the new individual, where the constraints are as follows. Determine convergence until the number of individuals selected reaches k, and calculate their probabilities and individual optimal solutions, where n b is the total number of branches; r i is the resistance of branch i; I i is the load current flowing through branch i; k i It is a 0-1 variable, representing the state of the branch, 0 means open, 1 means closed;
[0006] The constraints are as follows:
[0007] (1) Power flow equality constraint
[0008] The power flow equation that must be satisfied by network reconstruction is:
[0009]
[0010] Where: i = 1, 2, 3, ... n; k = 1, 2, 3, ... s; S i is the injected power of node i, P i and Q i are the injected active power and reactive power respectively; U i is the voltage vector of the node; Y ij is the admittance matrix element of the network node; n is the number of system nodes; s is the number of iterations.
[0011] (2) Node voltage constraints
[0012] The normal operation of the network has certain requirements for the power supply voltage. Under the premise of sufficient reactive power in the network, network reconstruction should increase the power supply voltage as much as possible.
[0013] Node Voltage Constraints
[0014]
[0015] Among them, |U i | is the voltage modulus of node i; U min and U max are the lower and upper limits of the node voltage respectively.
[0016] (3) Line capacity constraints
[0017] During the operation of the distribution network, in addition to requiring that the voltage amplitude of each node must change within the specified range, each branch must not be overloaded. Therefore, the following constraints are imposed on the line capacity:
[0018]
[0019] In the formula, |I j | is the actual current of line j, I jmax is the maximum current carrying capacity of line j.
[0020] (4) Connectivity
[0021] To ensure the normal power supply of all loads during the distribution network reconstruction, it is necessary to ensure that no islands appear in the reconstruction scheme. load The number of nodes N in the reconstruction scheme must meet the following conditions:
[0022] N=N load (4)
[0023] (5) Radial
[0024] The distribution network requires a closed-loop structure and open-loop operation. During normal operation, each node in the distribution network can only obtain power from one power source. Therefore, during normal operation, the distribution network supplies power in a radial manner. load Node N b The reconstruction scheme of the distribution network with branches must satisfy the following radial constraints while satisfying connectivity:
[0025] N load – N b = 1 (5)
[0026] ③ Determine whether the updated group is a radial network. If so, establish a distribution network flow calculation module and calculate network losses and node voltages. Otherwise, return to step ② to update the group;
[0027] ④ Establish a priori probability model, conditional probability model, and then establish a posterior probability model;
[0028] Because in the discrete optimization problem related to multiple variables, the value of each variable in the individual directly affects the individual fitness. The EDA algorithm relies on the probability of the variable value to find the best, therefore, establishing a probability model for the value of each variable in the fitness function is a key step in solving the problem using the EDA algorithm. The establishment and update mechanism of various probability models in this application are closely related to the Bayesian statistical inference theory. The probability models that need to be established include three models: prior probability, conditional probability and posterior probability.
[0029] (1) Establishment of prior probability model
[0030] In this paper, the individual variables in the EDA algorithm are encoded in binary. The variable value "0" represents the branch disconnection state, and "1" represents the closed state. Let the individual length be l, corresponding to the total number of branches in a distribution network. X = (x(1), x(2), ... x(l)), represents an individual, and x(j) is the variable at the jth position. The corresponding prior probability distribution is established for the corresponding variables at each position of the individual in the problem solution space. The prior probability distribution model of all individual variables is as follows:
[0031] q x =(q1,q2,…ql) T (6)
[0032] Prior probability is an important part of establishing and updating probability models. There are many ways to establish it, and you can choose a reasonable way based on the actual situation.
[0033] (2) Establishment of conditional probability model
[0034]
[0035] The value of each variable in an individual will affect other variables, so a conditional probability vector can be used to represent the degree of influence of the remaining variables on the variable xj. This conditional probability vector can be expressed as q j =(q(xj / x1),q(xj / x2),…,q(xj / xl)) T The above conditional probability can be obtained based on the individual information in the dominant population. It describes the conditional probability of variable xj when other variables take a certain value. Assume that the population evolves to the tth generation, an individual in the dominant population: X k t =(x k t (1),x k t (2),…x k t (l)), let the best individual X in this generation t* =(x t* (1),x t* (2),…x t* (l)). Using the prior probability distribution of the variable values in the dominant population and the optimal individual, the conditional probability distribution matrix of this generation is established As shown in formula (4-9):
[0036] From statistical theory we can get:
[0037] q t (x t* (j) / x t (i)) = q t(x t* (j)·x t (i)) / q t (x t (i)) (8)
[0038] In the formula, q t (x t* (j)·x t (i)) represents the probability that the variable xj in the winning individual and the variable xi in the dominant group take specific values at the same time, q t (x t (i)) represents the probability of variable xi taking a specific value in the dominant group.
[0039] (3) Establishment of the posterior probability model
[0040] As above, assuming that the population has evolved to the tth generation, the individual variable prior probability distribution and conditional probability distribution matrix Known. Then use the prior probability and conditional probability to establish the posterior probability model according to the Bayesian formula in:
[0041]
[0042] The denominator in the above formula is the product of the prior probability of the individual variable and the corresponding element in the conditional probability matrix of the variable x(j) and then the summation operation. It expresses the conditional probability of inferring the value of the remaining variables in the individual (the variable value is 0) under the condition that the variable of the optimal individual takes a specific value (here the branch is disconnected and the variable value is 0). The posterior probability matrix It combines the prior probability distribution with the excellent information of new individuals in the evolutionary process to provide better inference results.
[0043] ⑤ Sampling a new population from the posterior probability model and determining whether the termination condition is met. If so, jump back to the main program. Otherwise, select the dominant population, update the probability model, and sample a new population from the posterior probability. Termination condition: When the number of individuals in the initial population reaches the required value m, stop generating the initial population, and then hand over the generated initial population to the IEDA module for processing.
[0044] The updating of the probability model includes the following steps: first, the prior probability model and the conditional probability model are updated, and then the updated information of the two is combined to obtain the posterior probability model.
[0045] (1) Updating of prior probability model
[0046] Suppose the population evolves to the tth generation, and the winning individual is X t* =(x t* (1),xt* (2),…x t* (l)), taking the variable corresponding to the winning individual as the standard, combined with the advantage group D t The updated prior distribution probability of the individual is as follows:
[0047]
[0048] From the content of formula (11), we can know that δ j t (j=1,2,…l) The calculated value depends on the value of the variable at the same position of the dominant population and the optimal individual. When they are equal, the value is 1, and when they are not equal, the value is 0. α is a variable parameter greater than 0 and less than 1 in the Bayesian statistical inference probability update model. The larger the value of α, the greater the contribution to the updated prior probability distribution and the greater the impact on the subsequent population evolution direction. m is the number of individuals in the dominant population. The prior probability of each variable in the individual has a relationship of change in the same direction as the variable at the same position of the optimal individual taking the value of 0, that is, it increases when the value is 0. It decreases when the values are different. This method enhances the probability of the emergence and retention of good information and suppresses the transmission of bad information.
[0049] (2) Update of conditional probability model
[0050] The update of the conditional probability model is done through the dominant group D t The update method is as follows:
[0051]
[0052] In the formula It is an element in the conditional probability matrix model, a=b=1,2,…l, β∈(0,1), m parameter is the same as above, t is the evolutionary generation of the current dominant group, and the control parameter β has a similar effect to α. However, the control effect of β is bidirectional.
[0053] (3) Posterior Probability Model Update Method
[0054] By combining the updated prior probability model and the conditional probability model and calculating through the Bayesian formula, a new posterior probability model can be obtained.
[0055] The present invention also provides a distribution network reconstruction system, comprising:
[0056] The main program module defines the global structure variables and determines the population rules used for power flow calculation and execution of the IEDA algorithm.
[0057] The main program module includes the initial population generation function module, the network topology function module, the iterative calculation termination judgment function module, and the whole software interface building module. The function of the initial population generation function module is to randomly generate a single individual, and generate a specified number of initial populations through the connectivity and radial constraints of the distribution network; the function of the network topology function module is to verify whether the individuals randomly generated by the initial population generation function module meet the connectivity and radial constraints; the function of the iterative calculation termination judgment function module is to judge whether the entire algorithm has reached the end standard; the function of the whole software interface building module is to build an interactive user interface.
[0058] The power flow calculation module includes active network loss calculation and node voltage calculation. The specific calculation is as follows:
[0059] S i ' i =S ij +S im +S i (14)
[0060] After the forward calculation is completed, when performing the node voltage back calculation, it is assumed that the voltage U i It has been calculated that the voltage calculation formula of node j and node m is as follows:
[0061] BRIEF DESCRIPTION OF THE DRAWINGS
[0062] Figure 1 Radial distribution network with branch lines;
[0063] Figure 2 Node-branch relationship network;
[0064] Figure 3 Power flow calculation flow chart;
[0065] Figure 4 The initial structure of the local distribution network in Yantan area;
[0066] Figure 5 Voltage at each point before and after the reconstruction of the local distribution network in Yantan area;
[0067] Figure 6 The influence of control parameters on iteration speed;
[0068] Figure 7 The influence of control parameters on iteration speed;
[0069] Figure 8 The structure of the local distribution network in Yantan area after reconstruction. DETAILED DESCRIPTION
[0070] The present application is combined with the following embodiments to further describe the technical solution, but these embodiments are not intended to limit the scope of the present application.
[0071] Embodiment 1 A distribution network reconstruction system
[0072] A distribution network reconstruction system comprises:
[0073] The main program module, in order to define the global structural variables used in the flow calculation and the execution of the IEDA algorithm and determine the population rules, the main program module includes the initial population generation function module, the network topology function module, the iterative calculation termination judgment function module, and the whole software interface construction module. The function of the initial population generation function module is to randomly generate a single individual, and generate a specified number of initial populations through the connectivity and radial constraints of the distribution network; the function of the network topology function module is to verify whether the individuals randomly generated by the initial population generation function module meet the connectivity and radial constraints; the function of the iterative calculation termination judgment function module is to judge whether the entire algorithm has reached the end standard; the function of the whole software interface construction module is to build an interactive user interface.
[0074] The power flow calculation module includes active network loss calculation and node voltage calculation. The power flow calculation uses the forward-backward method, which has the following advantages: ① The calculation is time-saving and fast. Unlike the busbar method, it does not perform matrix operations. The forward method is based on Kirchhoff's theorem and does not need to calculate the power loss of the branch, which brings the advantage of calculation speed; ② Linear convergence, and the convergence characteristics are relatively good; ③ The hardware memory capacity required for the calculation is relatively small. Since the calculation process does not require network matrix calculation, all calculation processes and results can be stored only by vector calculation. The requirements for hardware storage devices are greatly reduced; ④ Due to the radial network topology and open-loop operation mode, the connection between various electrical parameters is simpler than that of the transmission network, so it is easy to program and implement, and is not sensitive to changes in network scale. Therefore, it has strong adaptability; ⑤ The requirements for the initial voltage value are not high, as long as it is near the rated value.
[0075] The forward and backward power flow calculation method of the distribution network is the same as the manual calculation method. First, use the known network node loads and node voltages to calculate the power loss of each branch and the injected power at the beginning and end of the branch from the end of the distribution network power flow to the beginning; then, based on the known power node voltage and the calculated power of each branch, calculate the voltage loss of each branch and the node voltage from the power node to the end node of the distribution network power flow. Figure 1It is a typical radial distribution network structure with branches. Like the network without branches, the entire network has only one power supply node (bus U1), which is still regarded as a balance node in the distribution network flow calculation. What is different from the network without branches is that each node of the trunk network has branch feeder lines and corresponding load nodes. The entire network structure is mapped to the graph theory model, and the relevant knowledge of graph theory is used to process the node-branch relationship of the distribution network flow calculation. This mapping is simple and easy to program. A breadth-first search is performed on the entire distribution network topology starting from the balance node to form a complete node-branch flow distribution directed relationship record. The balance node is usually used as the root node of the entire topology network, and branch nodes are generated from the root node to the outside, eventually forming a tree-like graph of the distribution network. Figure 1 In the example, the branch between nodes 2 and i is considered as a branching branch, and node i can also generate an outgoing branch. The branch between nodes i and m can be considered as a branching branch of node i. In addition, the branch between nodes i and m can also be considered as a sub-trunk branch, and the other branches are branching branches. The flow calculation method of the entire network is exactly the same as that of the unbranched distribution network. The only difference lies in the amount of data processing in the flow calculation. When calculating unbranched branches, there is only one child node under each parent node, and the amount of data calculation is relatively small. Some nodes with branched branches have more than one child node, and the amount of calculation is relatively large. For example, in Figure 1 In the calculation of the power received by node i, the power at the head end of the outgoing branch should be added. The specific calculation is:
[0076] S i ' i =S ij +S im +S i (19)
[0077] After the forward calculation is completed, when performing the node voltage back calculation, it is assumed that the voltage U i It has been calculated that the voltage calculation formula of node j and node m is as follows:
[0078]
[0079]
[0080] In order to speed up the calculation of power flow, this paper adopts the node-branch relationship model instead of the matrix arrangement in the calculation, such as Figure 2 As shown, after automatically reading the basic parameters of each node and branch of the distribution network, after selecting or determining the balancing node in the network according to the node type, the algebra of each node is arranged in order from small to large from the flow direction. Except for the balancing node and the flow end node, the remaining nodes have a parent node and a child node set. The data model is as follows:
[0081] Structure type: Node-branch relationship class Node_Branch{node number, parent node number, child node number set, node generation number, branch set}
[0082] The basic data model also includes: node basic data Node, original branch status Branch, and stored branch loss and node voltage data used in the later stage of calculation.
[0083] exist Figure 2 On the basis of node 0 being the balance node, a distribution network is formed in which the tidal current flows from the balance node to node 1. In this network, relative to node 2, node 1 is its parent node, and its child node set includes nodes 4 and 5. The node generation is calculated in ascending order along the tidal current direction with the balance node as the 0th generation. Node 2 is the 2nd generation, and node 3 also belongs to the second generation. The branch set it contains is a summary of branches named by the child node number. The branch set included in node 2 includes branch 4 and branch 5.
[0084] In the process of forward calculation of branch power flow, the network topology analysis is performed with width priority. To ensure forward calculation from the end of the power flow, the number of branches in the algebraic and branch set in the Node_Branch data type can ensure that the network changes and always forward calculation from the end of the power flow. The algebraic and branch set variable relationship in the Node_Branch data type naturally determines that each child node can perform power calculation in sequence. In the process of node voltage back-substitution calculation, the calculation is carried out from the balance node (root node) to each child node in sequence. After the back-substitution calculation is completed, the forward calculation is performed to prepare for the next node voltage back-substitution calculation. The calculation is repeated many times until the maximum value of the two most recent node voltage corrections meets the accuracy requirements. That is, Max(U[n+1]-U[n]<ε).
[0085] The distribution network topology analysis is performed by following the relationship between nodes and branches in the direction of the power flow, and then the power flow distribution is calculated. It has been proven that this node-branch relationship model can quickly perform network topology analysis, thereby improving the calculation speed. The calculation process is as follows Figure 3 shown.
[0086] Example 2 Testing the Applicability of the Reconfiguration Method to the Reconfiguration of the Structural Distribution Network
[0087] Part of the data of the 10kV distribution network in Yantan area, Chengguan District, Lanzhou City was intercepted to form a complete local distribution network, including 5 interconnecting branch switches, 34 nodes, 38 branches in total, the network nominal voltage is 10kV, the total load of the entire network is 11483.3kW+j7132kVar, and the power nodes are numbered 1, 2, and 3, respectively, which are respectively drawn from the 10kV side of the nearby 110kV substation. The network connection mode before reconstruction is the actual operation mode from January to October 2014, among which the interconnecting branches are 8-23, 10-11, 13-27, 20-26, and 21-29. At this time, the network active loss is 165.27kW. The calculation conditions are that the number of individuals in the initial population is s, which is 100, the dominant population selection rate M is 0.3, and the control parameters α and β are both 0.25. The initial network structure and the optimal structure after reconstruction are as follows Figure 4 , 5 shown.
[0088] ①Comparison of results before and after reconstruction
[0089] The relatively stable solutions (optimal spatial solutions) in the IEDA calculation process and optimization part are shown in Table 1 in the form of connecting branches. The 10 groups of optimal individuals in the table are completely retained in the last 5 iterations, and the optimal individuals appear continuously. The average number of iterations of the algorithm is 43, and the maximum number of repetitions of the optimal value is 8 times. The comparison of the results before and after reconstruction is shown in Tables 2 and Figure 5 As shown in Table 1, the interconnecting branches after reconstruction are: 13-14, 19-20, 22-23, 23-24, 10-11. Figure 5 It can be seen that through optimization, the active power loss of the network has been reduced by 58%, and the loss reduction effect is very significant. The voltage of the entire network has been greatly improved after reconstruction, and the voltage quality is generally high, which has been greatly improved overall compared with before reconstruction. The voltage at the lowest point of the network has increased from 0.9620pu to 0.9861pu, an increase of 2.5%. It can be seen that the optimization of the local network in Lanzhou area through the algorithm in this paper not only reduces the active power loss of the network, but also improves the voltage quality of the power supply, generating certain economic benefits.
[0090] Table 1 Optimal spatial solutions
[0091]
[0092] ②The influence of control parameters on algorithm performance
[0093] The probability update control factor selected by the IEDA algorithm has an impact on the algorithm's optimization speed. To further verify the impact of parameters α and β on the algorithm itself, the above calculation conditions are still used, that is, the number of individuals in the initial population s is still 100, the dominant population selection rate M is 0.3, the value of α is changed first, the β value remains unchanged, and then the value of β is changed, the α value remains unchanged, the optimal result is the same as before the parameter change, the top ten optimal solutions remain basically unchanged before and after the parameter change, and the solutions with lower fitness have slight changes before and after the parameter change, but the iterative speed of optimization is different under different α and β values, which leads to differences in the overall calculation speed of the algorithm. Taking the fitness of the optimal solution as a reference, the relationship between the degree of the algorithm's approach to the optimal space and the changes in parameters α and β and the number of iterations can be seen in Figure 6 , 7 The definition of “better space” is: a subset of the solution space consisting of a series of stable better solutions and the optimal solution that are closest to the optimal solution during the algorithm search process.
[0094] In this example, the optimal space is defined as the solution in Table 1.
[0095] Table 2 Comparison before and after network reconstruction
[0096] Table4.2 The comparison around network reconfiguration
[0097]
[0098] α is a variable parameter greater than 0 and less than 1 in the Bayesian statistical inference prior probability update model. Figure 6 From the influence curve of the parameter α on the number of algorithm iterations, it can be seen that when the value of α increases from 0.25 to 0.65, the number of iterations for the algorithm to enter the optimal space is smaller, that is, the algorithm calculation speed is faster. However, when α is greater than 0.73, as the value increases, the number of iterations increases very slowly. In summary, the larger the value of α, the greater the contribution to the updated prior probability distribution, and the greater the impact on the subsequent population evolution direction.
[0099] β is a variable parameter greater than 0 and less than 1 in the Bayesian statistical inference conditional probability update model. Figure 7From the influence curve of parameter β on the number of algorithm iterations, it can be seen that when the parameter value increases from 0.15 to 0.85, the number of iterations of the algorithm entering the optimal space changes in a bidirectional manner, that is, when the value increases from 0.15 to 0.61, the number of iterations of the algorithm entering the optimal space increases with the increase of the parameter. When the parameter β value exceeds 0.61, the number of algorithm iterations decreases with the increase of the parameter. In summary, the different values of β also show different trends in the influence of the updated conditional probability distribution. For a specific distribution network, the bidirectional demarcation point of parameter β is also different.
[0100] ③Since the application of EDA algorithm in distribution network reconstruction is still relatively rare, in order to further investigate the accuracy and calculation speed of the IEDA algorithm in this paper in the application of distribution network reconstruction, the particle swarm optimization algorithm (PSO) was used to reconstruct the distribution network. Then the simulation results were compared, and the comparison results are shown in Table 3. Under the condition that the initial population size is 100, the number of iterations of the two algorithms is not much different. The number of iterations of the PSO algorithm is more than that of the IEDA algorithm, and the average calculation time is also longer than that of the IEDA algorithm. The calculation results show that the optimal solution searched by the IEDA algorithm and the PSO algorithm is the same. Therefore, the accuracy and effectiveness of the IEDA algorithm in the application of Lanzhou distribution network reconstruction are verified by the simulation results. To a certain extent, the calculation speed of the IEDA algorithm is faster than that of the PSO.
[0101] Table 3 Comparison of reconstruction results of two algorithms
[0102]
Claims
1. A distribution network reconstruction method, It is characterized in that include: ① Establish the main program module and initialize the population: first close all the switch branches involved in the coding in the distribution network, and then randomly open a branch switch involved in the coding in each loop in turn until the total number of opened branches is equal to the total number of initial connecting switches in the network. Through this branch opening operation, an individual is generated, and then stored in the initial population after passing the inspection by the network topology analysis module in the main program. When the number of individuals in the initial population reaches the value m, stop generating the initial population; the value m is related to the specific scale of the distribution network. The larger the network scale, the larger the required value m, and the smaller the network scale, the smaller the value m. For a distribution network with n nodes and k connecting branches, the constraint condition is 0<m<C(n-1,k)-3C(n-2,k-1), where C(n-1,k) is the number of combinations; ② Update the population, establish the IEDA module, and set the population update rules: first determine the number of new individuals that the algorithm requires to be updated, and then sample the posterior probability model. The sampling method is based on each column of the updated posterior probability, and the dimension number with a larger probability of taking zero value for the corresponding dimension variable of the new individual is naturally selected until the number of zero-value variables selected is equal to the number of distribution network loops. The individuals are repeatedly sampled using the roulette method, and the new individuals and the current dominant population are combined into the next generation population. The formula is used Calculate the fitness of the new individual, where the constraints are as follows: (1) Power flow equality constraint The power flow equation that must be satisfied by network reconstruction is: Where: i = 1, 2, 3, ... n; k = 1, 2, 3, ... s; S i is the injected power of node i, P i and Q i are the injected active power and reactive power respectively; U i is the voltage vector of the node; Y ij is the admittance matrix element of the network node; n is the number of system nodes; s is the number of iterations; (2) Node voltage constraints The normal operation of the network has certain requirements for the power supply voltage. Under the premise of sufficient reactive power in the network, network reconstruction should increase the power supply voltage as much as possible. Node Voltage Constraints Among them, |U i | is the voltage modulus of node i; U min and U max are the lower and upper limits of the node voltage respectively; (3) Line capacity constraints During the operation of the distribution network, in addition to requiring that the voltage amplitude of each node must change within the specified range, each branch must not be overloaded. Therefore, the following constraints are imposed on the line capacity: In the formula, |I j | is the actual current of line j, I jmax is the maximum current carrying capacity of line j; (4) Connectivity To ensure the normal power supply of all loads during the distribution network reconstruction, it is necessary to ensure that no islands appear in the reconstruction scheme. load The number of nodes N in the reconstruction scheme must meet the following conditions: N=N load (5) Radial The distribution network requires a closed-loop structure to operate in an open-loop manner. During normal operation, each node in the distribution network can only obtain power from one power source. Therefore, during normal operation, the distribution network supplies power in a radial manner. load Node N b The reconstruction scheme of the distribution network with branches must satisfy the following radial constraints while satisfying connectivity: N load –N b =1 Determine the convergence until the number of individuals selected reaches k, and calculate their probabilities and individual optimal solutions, where n b is the total number of branches; r i is the resistance of branch i; I i is the load current flowing through branch i; k i It is a 0-1 variable, representing the state of the branch, 0 means open, 1 means closed; ③ Determine whether the updated group is a radial network. If so, establish a distribution network flow calculation module and calculate network losses and node voltages. Otherwise, return to step ② to update the group; ④ Establish a priori probability model, conditional probability model, and then establish a posterior probability model; ⑤ Sampling a new population from the posterior probability model and determining whether the termination condition is met. If so, jump back to the main program. Otherwise, select the dominant population, update the probability model, and sample a new population from the posterior probability. Termination condition: When the number of individuals in the initial population reaches the required value m, stop generating the initial population, and then hand over the generated initial population to the IEDA module for processing.
2. A distribution network reconstruction method as claimed in claim 1, It is characterized in that The establishment of the prior probability model: In the EDA algorithm, individual variables are encoded in binary, and the variable value "0" represents the branch disconnection state, and "1" represents the closed state. Assume that the individual length is l, corresponding to the total number of branches of a distribution network, X = (x(1), x(2), ... x(l)), represents an individual, x(j) is the variable at the jth position, and the corresponding prior probability distribution is established for the corresponding variables at each position of the individual in the problem solution space. The prior probability distribution model of all individual variables is as follows: <h2 style=";text-align:left;direction:ltr">q<h2 style=";text-align:left;direction:ltr"> x <h2 style=";text-align:left;direction:ltr"> (q1,q2,…ql)<h2 style=";text-align:left;direction:ltr"> T Prior probability is an important part of establishing and updating probability models. There are many ways to establish it, and you can choose a reasonable way based on the actual situation.
3. A distribution network reconstruction method as claimed in claim 1, It is characterized in that The establishment of the conditional probability model: A conditional probability vector is used to represent the influence of other variables on variable xj. This conditional probability vector is expressed as The above conditional probability can be obtained based on the individual information in the dominant population. It describes the conditional probability of variable xj when other variables take a certain value. Assuming that the population evolves to the tth generation, an individual in the dominant population: Suppose the best individual in this generation Using the prior probability distribution of the variable values in the dominant population and the optimal individual, the conditional probability distribution matrix of this generation is established. As shown in formula (4-9): From statistical theory we can get: In the formula, represents the probability that the variable xj in the winning individual and the variable xi in the dominant group take specific values at the same time, q t (x t (i)) represents the probability of variable xi taking a specific value in the dominant group.
4. A distribution network reconstruction method according to claim 1, It is characterized in that The establishment of the posterior probability model: Assuming that the population has evolved to the tth generation, the individual variable prior probability distribution and conditional probability distribution matrix If it is known, then use the prior probability and conditional probability to establish the posterior probability model according to the Bayesian formula in: The denominator in the above formula is the product of the prior probability of the individual variable and the corresponding element in the conditional probability matrix of the variable x(j), and then the sum is calculated. It expresses the conditional probability that the variable value of the optimal individual is 0 when the variable takes a specific value (here the branch is disconnected) and the variable value is 0.
5. A distribution network reconstruction method according to claim 1, It is characterized in that The updating method of the probability model comprises the following steps: firstly, updating the prior probability model and the conditional probability model, and then combining the updated information of the two to obtain the posterior probability model; (1) Updating the prior probability model: Assume that the population evolves to the tth generation and the winning individual is Taking the variable corresponding to the winning individual as the standard, combined with the advantage group D t The updated prior distribution probability of the individual is as follows: From the above formula, we can know that The calculated value depends on the value of the variable at the same position of the dominant population and the optimal individual. When they are equal, the value is 1, and when they are not equal, the value is 0. α is a variable parameter greater than 0 and less than 1 in the Bayesian statistical inference probability update model. The larger the value of α, the greater the contribution to the updated prior probability distribution and the greater the impact on the subsequent population evolution direction. m is the number of individuals in the dominant population. The prior probability of each variable in the individual has a relationship of change in the same direction as the variable at the same position of the optimal individual taking the value of 0, that is, it increases when the same value is 0, and decreases when the values are different. This method enhances the probability of the emergence and retention of good information and suppresses the transmission of bad information. (2) Update of conditional probability model: The update of conditional probability model should be done through the dominant group D t The update method is as follows: In the formula is an element in the conditional probability matrix model, a=b=1,2,…l, β∈(0,1), m parameter is the same as above, t is the evolutionary generation of the current dominant group, the effect of the control parameter β is similar to α, but the control influence of β is bidirectional; (3) Posterior probability model updating method: After combining the updated prior probability model and the conditional probability model and calculating through the Bayesian formula, a new posterior probability model can be obtained.
6. A distribution network reconstruction method according to any one of claims 1 to 5, It is characterized in that The distribution network reconstruction system includes: The main program module is used to define the global structural variables and determine the population rules for the execution of the power flow calculation and the IEDA algorithm; the main program module includes an initial population generation function module, a network topology function module, an iterative calculation termination judgment function module, and a complete set of software interface building modules; the function of the initial population generation function module is to randomly generate a single individual, and generate a specified number of initial populations through the connectivity and radial constraints of the distribution network; the function of the network topology function module is to verify whether the individuals randomly generated by the initial population generation function module meet the connectivity and radial constraints; the function of the iterative calculation termination judgment function module is to judge whether the entire algorithm has reached the end standard; the function of the complete set of software interface building modules is to build an interactive user interface; The power flow calculation module includes active network loss calculation and node voltage calculation. It first uses the known network node load and node voltage to calculate the power loss of each branch and the injected power at the beginning and end of the branch from the end of the distribution network power flow to the beginning; then, based on the known power node voltage and the calculated power of each branch, it calculates the voltage loss of each branch and the node voltage from the power node to the end node of the distribution network power flow.
