CN105069517A - Power distribution network multi-objective fault recovery method based on hybrid algorithm - Google Patents

Abstract

Novel hybrid fault recovery gaming comprises the following steps: 1) inputting an initial parameter, 2) establishing an initial matrix, 3) searching an initial solution set, 4) performing power verification, 5) performing networking correction, 6) performing first-class load division algorithm convergence inspection, 7) outputting a first-class load division result, 8) performing algorithm transition, 9) setting parameters of a quantum particle swarm algorithm, 10) initializing quantum particles, 11) performing objective function calculations, 12) updating the parameters, 13) updating position values and optimal vectors, 14) screening non-dominated solutions, 15) screening an elite set, 16) performing elimination operation, 17) performing quantum particle swarm algorithm convergence inspection, and 18) outputting a result. Through combinations with characteristics of a heuristic algorithm and an intelligent optimization algorithm, the hybrid fault recovery method is provided, and a fault recovery problem of a power distribution network containing DG is solved.

Description

Based on the power distribution network multiple goal fault recovery method of hybrid algorithm

Technical field

The present invention relates to a kind of in conjunction with heuristic and novel mixed fault restoration methods that is intelligent optimization algorithm, especially for a kind of power distribution network multiple goal fault recovery problem containing distributed power source considering load level.

Background technology

Along with the reinforcement of distribution network construction and reaching its maturity of micro-capacitance sensor technology, the fault recovery of power distribution network becomes the important step of power grid self-healing control gradually.Along with distributed power source (distributedgeneration, DG) power distribution network is accessed in a large number, traditional single source radial distribution networks becomes multi-source system, electric network composition is complicated all the more, if rational recovery policy cannot be formulated rapidly for fault, realize each type load service restoration of power distribution network, distributed power source and power distribution network will bear massive losses, and the safe and stable operation of whole power distribution network will be had a strong impact on.

Fault recovery containing distributed power source power distribution network is a kind of extensive, non-linear, multiobject combinatorial optimization problem, and method for solving main at present mainly contains heuristic search and intelligent optimization method.Heuristic search is by formulating corresponding heuristic search rule, and obtain fail-over path, search speed is fast.But, above proposed method all belongs to the two-step optimization method of first search, rear adjustment, the rule of adjustment is artificially formulated usually, due to the limitation of artificial experience, the formulation of heuristic rule is often more difficult and not comprehensive, easily make final optimum results be absorbed in local optimum, algorithm lacks applicability widely.And when using heuritic approach to solve multi-objective optimization question, often need to be translated into single-object problem by weight factor and solve, make the inefficiency obtaining optimum solution.Intelligent optimization method is applied to fault recovery multi-objective problem mainly through particle cluster algorithm, genetic algorithm, evolution algorithm etc. and solves.Because intelligent algorithm adopts the mode of random optimizing, in optimizing process, easily produce the infeasible solution running counter to the radial constraint of power distribution network, power-balance constraint in a large number, if do not revise, the efficiency of algorithm will reduce, and is easily absorbed in locally optimal solution.Therefore, the multiple goal fault recovery problem containing distributed power source can not all be processed preferably only by heuristic search or intelligent optimization method.

Summary of the invention

The present invention will overcome the above-mentioned shortcoming of prior art, proposes a kind of in conjunction with heuristic and novel mixed fault restoration methods that is intelligent optimization algorithm.

Load on society, the economic impact produced, is divided into a class, two classes and three type loads according to load by electric system.Wherein, a type load all must keep the load of power supply under referring in office why barrier.At present, for considering that the distribution network failure of load level recovers problem, adopt heuritic approach and intelligent optimization algorithm, all can not obtain good effect: the distribution network failure solving consideration three type load grade according to heuritic approach recovers problem, heuristic rule is formulated very complicated, and the more difficult consideration of the formulation of rule is thorough, document is rarely had to relate at present; According to intelligent optimization algorithm, although without the need to formulating corresponding heuristic rule, need to revise the infeasible solution in iterative process, real-time is not strong, can only obtain local solution when solving multiple constraint and multivariable problem; And final being difficult to ensures that all type loads restore electricity completely.Therefore, fault recovery PROBLEM DECOMPOSITION containing DG power distribution network is that a type load divides and system failure recovery reconstructs two subproblems by project of the present invention, the mixed fault restoration methods that a kind of heuritic approach and intelligent optimization algorithm combine is proposed: wherein, subproblem is divided for a type load, heuristic is adopted to solve, to ensure that in final fail-over policy, a type load can restore electricity completely; For fault recovery reconstruct subproblem, intelligent optimization algorithm is adopted to solve, minimum to ensure weighting dead electricity load.

Project of the present invention adopt the mixed fault restoration methods based on heuritic approach and quanta particle swarm optimization to solve to consider load level containing DG distribution network failure Restoration model, concrete Optimizing Flow is as shown in Figure 1.Power distribution network multiple goal fault recovery method based on hybrid algorithm of the present invention, step is as described below:

1) initial parameter is inputted

The generation position of input fault, the design parameter of power distribution network, comprise micro-source number N _dG,, power distribution network master switch number N _b, load level parameter;

2) initial matrix is set up

For realizing the division of a type load, the inventive method defines four kinds of initial matrixs for heuristic search algorithm, is respectively load-micro-source scaling matrices, load-micro-source ordinal matrix, load-micro-source ownership matrix and Wei Yuan-micro-source networking matrix.Each defined matrix is as described below:

2.1) load-micro-source scaling matrices

Micro-source-load proportion matrix A _peraccount for the percent information of this micro-source capacity for recording in power distribution network all load total amounts on a type load to each Wei Yuan UNICOM branch road.At A _perin matrix, a type load corresponding row vector, micro-source respective column vector, then on element representation load i to the Wei Yuan j UNICOM branch road of matrix i-th row jth row, all load total amounts account for the percent information of micro-source j capacity, and expression formula is:

$A_{per(i,j)}=\frac{\underset{z\∈L_{i,j}}{\Σ}{P}_{LD,z}}{P_{DG,j}}---\left(1\right)$

Wherein, L _i,jexpress the load aggregation on load i and the shortest communication path of micro-source j, P _{lD, z}for the power of load z on all communication paths, P _{dG, j}for the active power of micro-source j.A _{per (i, j)}numerical value is less, represents that load i is more likely divided to micro-source j.

2.2) load-micro-source ordinal matrix

Load-micro-source ordinal matrix A _sorit is the sequencing information accounting for this micro-source capacity ratio for recording a type load in power distribution network to all load total amounts on each Wei Yuan UNICOM branch road.Wherein, a type load numbering corresponding row vector, micro-source sequence respective column vector, the then A corresponding to element representation of matrix i-th row jth row _perelements A in matrix _{per (i, j)}in the sequence of the i-th row, sorting arranges from small to large by numerical values recited.Such as A _perelements A in matrix _{per (2,2)}corresponding numerical value is 1.2, after its place row sorts from small to large, is positioned at first, so A _sora corresponding in matrix _{sor (2,2)}the numerical value of element is 1.

2.3) load-micro-source ownership matrix

Load-micro-source ownership matrix A _belrepresent according to certain rule, in power distribution network, a type load is divided to the information in certain micro-source.Wherein, a type load numbering corresponding row vector, micro-source numbering respective column vector, in matrix, the i-th row, j column element are 1, represent load i and belong to micro-source j, and 0 represents that load i does not belong to micro-source j.

2.4) micro-source-micro-source networking matrix

When certain micro-source cannot meet the larger type load of one or more capacity, carry out networking between Wei Yuan and micro-source to meet the larger type load of capacity.Networking matrix A _unirow vector and column vector all represent that micro-source is numbered, if the i-th row, jth column element are 1, represent that networking is carried out in an i-th micro-source and jth micro-source, form an island network system.In the algorithm starting stage, each self-forming in each micro-source network, therefore, networking matrix is unit matrix.

3) initial disaggregation is searched for

For realizing the division of a type load, needing to formulate relevant rule set and being used for solving of heuristic search algorithm.The inventive method proposes 2 kinds of heuristic rules for searching for initial disaggregation, and rule is described below:

Rule 1: in a final type load division result, a type load i can not belong to A _perin matrix i-th row, numerical value is more than or equal to micro-source of 1.Because, if A _perin matrix, an i-th row jth element is more than or equal to 1, shows that load i is greater than micro-source j capacity to the load total amount on the j communication path of micro-source, does not meet constraint condition (2), i.e. formula (6).

Rule 2: compared to A _sorin the larger micro-source j of the i-th line ordering sequence number, a type load i more easily belong to the less micro-source j of sequence sequence number ^*.Because sequence sequence number is little, show that a type load i is to micro-source j ^*it is less that load total amount on communication path accounts for micro-source capacity ratio, and this micro-source can hold a more type load.

According to above-mentioned heuristic rule, the search step of initial network disaggregation is described below: first, according to topology of networks after distribution network failure, draws load-micro-source scaling matrices A _per, load-micro-source ordinal matrix A _sor.Secondly, according to regular 1 Sum fanction 2, draw load-micro-source ownership matrix A _bel.Finally, according to matrix A _bel, the initial division result of a type load after intelligent distribution network fault can be drawn.

4) power verification

For realizing the power verification of each island network, project of the present invention proposes following heuristic rule:

Rule 3: the island network at all Wei Yuan places must meet power-balance verification, namely the capacity in this micro-source must be more than or equal to the power summation of all loads on all type loads of this Wei Yuan place network and a type load to micro-source communication path.Check formula is:

$\underset{z\∈D_j}{\Σ}{P}_z+P_{loss,j}\≤P_{DG,j}---\left(2\right)$

Wherein, P _zrepresent the power of load z, set D _jrepresent that type loads all in the island network at j place, micro-source and all type loads are to all loads on the j communication path of micro-source; P _{loss, j}represent the network loss of isolated island by obtaining after Load flow calculation residing for micro-source j.

Rule 4: if micro-source j does not meet power-balance verification, then reject A corresponding to a type load in micro-source j _perthe load that in matrix, numerical value is maximum, and this load is divided to its A _sormicro-source of the rear cis-position of corresponding row sequence in matrix.

According to above-mentioned heuristic rule, the step of power verification is as described below: for the splitting scheme of a type load initial in step (3), regularly 3 carry out power verification to the network at each Wei Yuan place, all micro-sources all verify one and take turns and be designated as an iteration.If in an iteration, all micro-sources all meet rule 3, then enter step 6), if do not meet, then regularly 4 revise, upgrade matrix A _bel, then carry out next round iteration, until all micro-sources all meet rule 3.If certain load is repeatedly rejected by micro-source, illustrate that single micro-source cannot carry this load, then enter networking correction step, i.e. step 5).

5) networking correction

If certain load is repeatedly rejected by micro-source, illustrate that single micro-source cannot carry this load, need networking operation be carried out.For realizing the networking correction between each micro-source, the inventive method proposes following heuristic rule:

Rule 5: if 2 to be divided in any micro-source by a type load i according to rule, the network at this Wei Yuan place all can be caused not meet formula (2), then by A corresponding to this load _sortwo that sort the most forward in matrix i-th row micro-source composition networkings, this type load is divided in this networking accordingly; If after inspection, still do not meet formula (2), then first three micro-source should form networking, the like.

According to above-mentioned heuristic rule, the step of networking correction is as follows: record the number of times that each load is disallowable in each iterative process, if the number of times disallowable in this iteration of certain load exceedes micro-source sum, then carry out networking correction according to rule 5, simultaneously and upgrade matrix A _per, A _beland A _uni, and be back to power checking procedure and carry out next round iteration, successively each networking is verified.

6) a type load partitioning algorithm convergence inspection

Judge whether that all micro-sources all meet formula (2), if meet, then enter step (7), if do not meet, get back to step (4), continue to carry out power verification to the network at each Wei Yuan place, carry out next round iteration.

7) a type load division result is exported

Heuritic approach terminates, and exports a type load division result, obtains the switch sequence number set closed.

8) algorithm transition

According to closed switch sequence number set, to not determining that the switch-linear hybrid of on off state is the optimized variable of quantum discrete particle cluster algorithm.

9) quanta particle swarm optimization optimum configurations

The dimension of quantum discrete particle cluster algorithm, iterations and corresponding parameter value are set.

10) initialization of quanta particle

The positional value x of each particle of initialization ^kelite's collection of (i.e. the quantity of state of switch), quantum bit position, rotation angle, local optimum vector x p and non-domination solution, picks out non-domination solution that elite concentrates crowding distance minimum as global optimum vector x g.

11) objective function calculates

According to the positional value of each particle, draw the running status of power distribution network in conjunction with a type load division result, and according to objective function (3) and (4), calculate the adaptive value of each particle.

12) parameter upgrades

According to the more new formula of quantum particle swarm, upgrade the bit of quantum rotation angle guiding value, quantum rotation angle and quanta particle successively.

13) positional value and optimal vector upgrade

According to the more new formula of quantum particle swarm, upgrade positional value x ^k, and local optimal vector.

14) non-domination solution screening

According to the target function value of each particle, filter out Pareto optimum solution, and it is collected stored in elite.

15) elite collects screening

Current population is carried out to the computing of non-domination solution, the non-domination solution finally obtained is put into elite's collection.Select non-domination solution that elite concentrates the spacing of individuality maximum as globally optimal solution.

16) computing is eliminated

Utilize and improve microhabitat filtering technique, carry out superseded computing to the non-domination solution that elite concentrates, elite concentrates the diversity of population.

17) quanta particle swarm optimization test for convergence.

Whether check algorithm restrains.If so, then step 18 is entered); If not, then get back to step 11).

18) result exports

According to the net result that heuritic approach and quanta particle swarm optimization draw, the fail-over policy of output distribution net.

Advantage of the present invention is: the inventive method can reduce a large amount of infeasible solutions directly adopting and produce in intelligent optimization algorithm process, and principle is simple, be applicable to consider recovering containing DG distribution network failure of load level, make a type load in final fault recovery scheme can ensure to power completely.

Accompanying drawing explanation

Fig. 1 is algorithm flow of the present invention

Fig. 2 is the network structure of example of the present invention

Fig. 3 is all previous iteration result of three matroids of the present invention, and wherein, Fig. 3 a is A _permatrix, Fig. 3 b is A _belmatrix, Fig. 3 c is A _unimatrix

Fig. 4 is the final network structure that a type load of the present invention divides

Fig. 5 is the Pareto optimality face that two class algorithms of the present invention draw

Embodiment

Below in conjunction with accompanying drawing, further illustrate technical scheme of the present invention.

1. technical scheme of the present invention.

The DG distribution network failure that contains based on hybrid algorithm of the present invention recovers to put method, and step is as described below:

1) initial parameter is inputted

The generation position of input fault, the design parameter of power distribution network, comprise micro-source number N _dG,, power distribution network master switch number N _b, load level parameter;

2) initial matrix is set up

For realizing the division of a type load, the inventive method defines four kinds of initial matrixs for heuristic search algorithm, is respectively load-micro-source scaling matrices, load-micro-source ordinal matrix, load-micro-source ownership matrix and Wei Yuan-micro-source networking matrix.Each defined matrix is as described below:

2.1) load-micro-source scaling matrices

Micro-source-load proportion matrix A _peraccount for the percent information of this micro-source capacity for recording in power distribution network all load total amounts on a type load to each Wei Yuan UNICOM branch road.At A _perin matrix, a type load corresponding row vector, micro-source respective column vector, then on element representation load i to the Wei Yuan j UNICOM branch road of matrix i-th row jth row, all load total amounts account for the percent information of micro-source j capacity, and expression formula is:

$A_{per(i,j)}=\frac{\underset{z\∈L_{i,j}}{\Σ}{P}_{LD,z}}{P_{DG,j}}---\left(1\right)$

Wherein, L _i,jexpress the load aggregation on load i and the shortest communication path of micro-source j, P _{lD, z}for the power of load z on all communication paths, P _{lD, j}for the active power of micro-source j.A _{per (i, j)}numerical value is less, represents that load i is more likely divided to micro-source j.

2.2) load-micro-source ordinal matrix

Load-micro-source ordinal matrix A _sorit is the sequencing information accounting for this micro-source capacity ratio for recording a type load in power distribution network to all load total amounts on each Wei Yuan UNICOM branch road.Wherein, a type load numbering corresponding row vector, micro-source sequence respective column vector, the then A corresponding to element representation of matrix i-th row jth row _perelements A in matrix _{per (i, j)}in the sequence of the i-th row, sorting arranges from small to large by numerical values recited.Such as A _perelements A in matrix _{per (2,2)}corresponding numerical value is 1.2, after its place row sorts from small to large, is positioned at first, so A _sora corresponding in matrix _{sor (2,2)}the numerical value of element is 1.

2.3) load-micro-source ownership matrix

Load-micro-source ownership matrix A _belrepresent according to certain rule, in power distribution network, a type load is divided to the information in certain micro-source.Wherein, a type load numbering corresponding row vector, micro-source numbering respective column vector, in matrix, the i-th row, j column element are 1, represent load i and belong to micro-source j, and 0 represents that load i does not belong to micro-source j.

2.4) micro-source-micro-source networking matrix

When certain micro-source cannot meet the larger type load of one or more capacity, carry out networking between Wei Yuan and micro-source to meet the larger type load of capacity.Networking matrix A _unirow vector and column vector all represent that micro-source is numbered, if the i-th row, jth column element are 1, represent that networking is carried out in an i-th micro-source and jth micro-source, form an island network system.In the algorithm starting stage, each self-forming in each micro-source network, therefore, networking matrix is unit matrix.

3) initial disaggregation is searched for

For realizing the division of a type load, needing to formulate relevant rule set and being used for solving of heuristic search algorithm.The inventive method proposes 2 kinds of heuristic rules for searching for initial disaggregation, and rule is described below:

Rule 1: in a final type load division result, a type load i can not belong to A _perin matrix i-th row, numerical value is more than or equal to micro-source of 1.Because, if A _perin matrix, an i-th row jth element is more than or equal to 1, shows that load i is greater than micro-source j capacity to the load total amount on the j communication path of micro-source, does not meet constraint condition (2), i.e. formula (6).

Rule 2: compared to A _sorin the larger micro-source j of the i-th line ordering sequence number, a type load i more easily belong to the less micro-source j of sequence sequence number ^*.Because sequence sequence number is little, show that a type load i is to micro-source j ^*it is less that load total amount on communication path accounts for micro-source capacity ratio, and this micro-source can hold a more type load.

According to above-mentioned heuristic rule, the search step of initial network disaggregation is described below: first, according to topology of networks after distribution network failure, draws load-micro-source scaling matrices A _per, load-micro-source ordinal matrix A _sor.Secondly, according to regular 1 Sum fanction 2, draw load-micro-source ownership matrix A _bel.Finally, according to matrix A _bel, the initial division result of a type load after intelligent distribution network fault can be drawn.

4) power verification

For realizing the power verification of each island network, project of the present invention proposes following heuristic rule:

Rule 3: the island network at all Wei Yuan places must meet power-balance verification, namely the capacity in this micro-source must be more than or equal to the power summation of all loads on all type loads of this Wei Yuan place network and a type load to micro-source communication path.Check formula is:

$\underset{z\∈D_j}{\Σ}{P}_z+P_{loss,j}\≤P_{DG,j}---\left(2\right)$

Wherein, P _zrepresent the power of load z, set D _jrepresent that type loads all in the island network at j place, micro-source and all type loads are to all loads on the j communication path of micro-source; P _{loss, j}represent the network loss of isolated island by obtaining after Load flow calculation residing for micro-source j.

Rule 4: if micro-source j does not meet power-balance verification, then reject A corresponding to a type load in micro-source j _perthe load that in matrix, numerical value is maximum, and this load is divided to its A _sormicro-source of the rear cis-position of corresponding row sequence in matrix.

According to above-mentioned heuristic rule, the step of power verification is as described below: for the splitting scheme of a type load initial in step (3), regularly 3 carry out power verification to the network at each Wei Yuan place, all micro-sources all verify one and take turns and be designated as an iteration.If in an iteration, all micro-sources all meet rule 3, then enter step 6), if do not meet, then regularly 4 revise, upgrade matrix A _bel, then carry out next round iteration, until all micro-sources all meet rule 3.If certain load is repeatedly rejected by micro-source, illustrate that single micro-source cannot carry this load, then enter networking correction step.

5) networking correction

If certain load is repeatedly rejected by micro-source, illustrate that single micro-source cannot carry this load, need networking operation be carried out.For realizing the networking correction between each micro-source, the inventive method proposes following heuristic rule:

Rule 5: if 2 to be divided in any micro-source by a type load i according to rule, the network at this Wei Yuan place all can be caused not meet formula (2), then by A corresponding to this load _sortwo that sort the most forward in matrix i-th row micro-source composition networkings, this type load is divided in this networking accordingly; If after inspection, still do not meet formula (2), then first three micro-source should form networking, the like.

According to above-mentioned heuristic rule, the step of networking correction is as follows: record the number of times that each load is disallowable in each iterative process, if the number of times disallowable in this iteration of certain load exceedes micro-source sum, then carry out networking correction according to rule 5, simultaneously and upgrade matrix A _per, A _beland A _uni, and be back to power checking procedure and carry out next round iteration, successively each networking is verified.

6) a type load partitioning algorithm convergence inspection

Judge whether that all micro-sources all meet formula (2), if meet, then enter step (7), if do not meet, get back to step (4), continue to carry out power verification to the network at each Wei Yuan place, carry out next round iteration.

7) a type load division result is exported

Heuritic approach terminates, and exports a type load division result, obtains the switch sequence number set closed.

8) algorithm transition

According to closed switch sequence number set, to not determining that the switch-linear hybrid of on off state is the optimized variable of quantum discrete particle cluster algorithm.

9) quanta particle swarm optimization optimum configurations

The dimension of quantum discrete particle cluster algorithm, iterations and corresponding parameter value are set.

10) initialization of quanta particle

The positional value x of each particle of initialization ^kelite's collection of (i.e. the quantity of state of switch), quantum bit position, rotation angle, local optimum vector x p and non-domination solution, picks out non-domination solution that elite concentrates crowding distance minimum as global optimum vector x g.

11) objective function calculates

According to the positional value of each particle, draw the running status of power distribution network in conjunction with a type load division result, and according to objective function (3) and (4), calculate the adaptive value of each particle.

12) parameter upgrades

According to the more new formula of quantum particle swarm, upgrade the bit of quantum rotation angle guiding value, quantum rotation angle and quanta particle successively.

13) positional value and optimal vector upgrade

According to the more new formula of quantum particle swarm, upgrade positional value x ^k, and local optimal vector.

14) non-domination solution screening

According to the target function value of each particle, filter out Pareto optimum solution, and it is collected stored in elite.

15) elite collects screening

Current population is carried out to the computing of non-domination solution, the non-domination solution finally obtained is put into elite's collection.Select non-domination solution that elite concentrates the spacing of individuality maximum as globally optimal solution.

16) computing is eliminated

Utilize and improve microhabitat filtering technique, carry out superseded computing to the non-domination solution that elite concentrates, elite concentrates the diversity of population.

17) quanta particle swarm optimization test for convergence.

Whether check algorithm restrains.If so, then step 18 is entered); If not, then get back to step 11).

18) result exports

According to the net result that heuritic approach and quanta particle swarm optimization draw, the fail-over policy of output distribution net.

2. the embodiment of project of the present invention

Fault recovery PROBLEM DECOMPOSITION containing DG power distribution network is that a type load divides and system failure recovery reconstructs two subproblems by project of the present invention, the mixed fault restoration methods that a kind of heuritic approach and intelligent optimization algorithm combine is proposed: wherein, subproblem is divided for a type load, heuristic is adopted to solve, to ensure that in final fail-over policy, a type load can restore electricity completely; For fault recovery reconstruct subproblem, intelligent optimization algorithm is adopted to solve, minimum to ensure weighting dead electricity load.Concrete Optimizing Flow as shown in Figure 1.Solution procedure is as described below.

1) initial parameter is inputted

The generation position of input fault, the design parameter of power distribution network, comprise micro-source number N _dG,, power distribution network master switch number N _b, load level parameter;

2) initial matrix is set up

For realizing the division of a type load, the inventive method defines four kinds of initial matrixs for heuristic search algorithm, is respectively load-micro-source scaling matrices, load-micro-source ordinal matrix, load-micro-source ownership matrix and Wei Yuan-micro-source networking matrix.Self-definedly initial matrix is set up according to each.

3) initial disaggregation is searched for

First, according to topology of networks after distribution network failure, draw load-micro-source scaling matrices A _per, load-micro-source ordinal matrix A _sor.Secondly, according to regular 1 Sum fanction 2, draw load-micro-source ownership matrix A _bel.Finally, according to matrix A _bel, the initial division result of a type load after intelligent distribution network fault can be drawn.

4) power verification

For the splitting scheme of a type load initial in step (3), regularly 3 carry out power verification to the network at each Wei Yuan place, all micro-sources all verify one and take turns and be designated as an iteration.If in an iteration, all micro-sources all meet rule 3, then enter step 6), if do not meet, then regularly 4 revise, upgrade matrix A _bel, then carry out next round iteration, until all micro-sources all meet rule 3.If certain load is repeatedly rejected by micro-source, illustrate that single micro-source cannot carry this load, then enter networking correction step.

5) networking correction

Record the number of times that each load is disallowable in each iterative process, if the number of times disallowable in this iteration of certain load exceedes micro-source sum, then carry out networking correction according to rule 5, simultaneously and upgrade matrix A _per, A _beland A _uni, and be back to power checking procedure and carry out next round iteration, successively each networking is verified.

6) a type load partitioning algorithm convergence inspection

Judge whether that all micro-sources all meet formula (2), if meet, then enter step (7), if do not meet, get back to step (4), continue to carry out power verification to the network at each Wei Yuan place, carry out next round iteration.

7) a type load division result is exported

Heuritic approach terminates, and exports a type load division result, obtains the switch sequence number set closed.

8) algorithm transition

According to closed switch sequence number set, to not determining that the switch-linear hybrid of on off state is the optimized variable of quantum discrete particle cluster algorithm.

9) quanta particle swarm optimization optimum configurations

The dimension of quantum discrete particle cluster algorithm, iterations and corresponding parameter value are set.

10) initialization of quanta particle

The positional value x of each particle of initialization ^kelite's collection of (i.e. the quantity of state of switch), quantum bit position, rotation angle, local optimum vector x p and non-domination solution, picks out non-domination solution that elite concentrates crowding distance minimum as global optimum vector x g.

11) objective function calculates

According to the positional value of each particle, draw the running status of power distribution network in conjunction with a type load division result, and according to objective function (3) and (4), calculate the adaptive value of each particle.

12) parameter upgrades

According to the more new formula of quantum particle swarm, upgrade the bit of quantum rotation angle guiding value, quantum rotation angle and quanta particle successively.

13) positional value and optimal vector upgrade

According to the more new formula of quantum particle swarm, upgrade positional value x ^k, and local optimal vector.

14) non-domination solution screening

According to the target function value of each particle, filter out Pareto optimum solution, and it is collected stored in elite.

15) elite collects screening

Current population is carried out to the computing of non-domination solution, the non-domination solution finally obtained is put into elite's collection.Select non-domination solution that elite concentrates the spacing of individuality maximum as globally optimal solution.

16) computing is eliminated

Utilize and improve microhabitat filtering technique, carry out superseded computing to the non-domination solution that elite concentrates, elite concentrates the diversity of population.

17) quanta particle swarm optimization test for convergence.

Whether check algorithm restrains.If so, then step 18 is entered); If not, then get back to step 11).

18) result exports

According to the net result that heuritic approach and quanta particle swarm optimization draw, the fail-over policy of output distribution net.

3. case analysis

For verifying rationality and the practicality of this project, verified by following example.Project of the present invention adopts the IEEE33 node system improved to be example, and the topological structure of its distribution system as shown in Figure 2.This distribution system comprises 5 interconnection switches, and load total amount is 3715kW, and load level optimum configurations is as shown in table 1;

Table 1

Table 2

5 the DG parameters introduced are as shown in table 2, and node type is set to PQ node.Suppose that circuit (1) place breaks down, interpretation of result is as described below.

3.1) a type load partitioning algorithm interpretation of result

1) utilize heuritic approach to carry out the division of a type load, obtain the division result of a type load, algorithm is restrained after 2 iteration.Fig. 3 illustrates the load-micro-source scaling matrices A in all previous iterative process _per, load micro-source ownership matrix A _belwith micro-source-micro-source networking matrix A _uniresult.As shown in Figure 3, be positioned at the initial solution of three matrix representation heuritic approach starting stages of first row, secondary series and the 3rd column matrix represent respectively first time iteration and second time iteration after result.Table 3 illustrates the type load division result in all previous iterative process of heuritic approach; Sequence number in table is divided to node (load) sequence number in certain micro-source when representing all previous iteration.Fig. 4 is illustrated the type load obtained by heuritic approach and divides net result.

Table 3

2) three matroid initial solutions are as shown in the first row matrix in Fig. 3.Observe A _perthe initial results of matrix, in node 30 and node 32 corresponding row, the numerical value of each element is all greater than 1, does not meet rule 1.Therefore, in lower whorl iteration, repartition to the load of node 30 and node 32, for it mates new micro-source, or networking correction is carried out in corresponding micro-source.

3) three matroids first time iteration result as shown in the secondary series matrix in Fig. 3.Observe second A in Fig. 3 (c) _unimatrix, DG2 and DG5 forms networking, and according to formula (13), the 2nd row of this matrix are identical with the 5th column element numerical value.Observe second A in Fig. 3 (a) _permatrix, each element numerical value of node 30 corresponding row is all greater than 1, does not meet rule 1, and algorithm continues iteration.

4) the second time iteration result of three matroids is as shown in the 3rd column matrix in Fig. 3.By A _unithe second time iteration result of matrix is known, and DG1, DG2 and DG5 form networking; Observe A _perthe second time iteration result of matrix, in matrix, the minimum value of every row element numerical value is all less than 1, meets rule 1.According to power-balance inspection formula (2), all meet power-balance constraint in each networking, algorithm iteration terminates.A final type load division result as shown in Figure 4, in figure sequence number be 3,4,9,15-20,22,30-32,34 and 36 represents the switch sequence number of the closure state obtained by heuritic approach.Visible, by a type load partitioning algorithm, can be met quickly a type load power constraint initial division result, making the system failure recovery restructing algorithm of second step without the need to considering this constraint condition, contributing to rapid solving; Meanwhile, determine the state of partial switch variable, reduce the dimension of variable to be solved, reduce in intelligent optimization algorithm the possibility producing a large amount of infeasible solution.

3.2) multiple-objection optimization result

1) result obtained is divided for a type load, for remainder, quantum particle swarm multi-objective optimization algorithm only need not determine that multiple-objection optimization calculating is carried out in state of switch sequence number and constraint condition (2)-(5), greatly reduce the quantity solving difficulty and infeasible solution.The switch sequence number solved is 2,5-8,10-14,21,23-29,33,35 and 37, amount to 21 switches.Therefore, the dimension arranging each particle of multi-objective particle swarm algorithm is 21, and population is 500, iterations sum 2000 times.

2) the Pareto optimum solution of trying to achieve based on the fault recovery method of hybrid algorithm and fault recovery scheme as shown in table 4.In table, scheme 1 is relative less with scheme 2 switching manipulation number of times, scheme 3 and scheme 4 weighting load loss less.Therefore, when determining final plan, different schemes can be selected according to corresponding demand.The Pareto optimum solution obtained based on traditional quanta particle swarm optimization and fault recovery scheme as shown in table 5.Can find out, compared to the result that hybrid algorithm is tried to achieve, when same switch action frequency, in the scheme that traditional quanta particle swarm optimization obtains, weighting load loss is comparatively large, and scheme is poor.

Table 4

Table 5

3) the optimum face of the Bi-objective Pareto tried to achieve based on hybrid algorithm and traditional quanta particle swarm optimization as shown in Figure 5.As seen from the figure, context of methods is owing to have employed a type load division methods early stage, the optimum face of the Pareto of final acquisition relative to traditional quantum particle swarm obtain better, and the number of optimum solution is also many compared with traditional quantum particle swarm, in the application process of reality, more how more efficiently strategy can be provided to scheduler.

Claims (1)

1., based on the power distribution network multiple goal fault recovery method of hybrid algorithm, step is as described below:

1) initial parameter is inputted;

The generation position of input fault, the design parameter of power distribution network, comprise micro-source number N _dG, power distribution network master switch number N _b, load level parameter;

2) initial matrix is set up;

For realizing the division of a type load, the inventive method defines four kinds of initial matrixs for heuristic search algorithm, is respectively load-micro-source scaling matrices, load-micro-source ordinal matrix, load-micro-source ownership matrix and Wei Yuan-micro-source networking matrix; Each defined matrix is as described below:

2.1) load-micro-source scaling matrices;

Micro-source-load proportion matrix A _peraccount for the percent information of this micro-source capacity for recording in power distribution network all load total amounts on a type load to each Wei Yuan UNICOM branch road; At A _perin matrix, a type load corresponding row vector, micro-source respective column vector, then on element representation load i to the Wei Yuan j UNICOM branch road of matrix i-th row jth row, all load total amounts account for the percent information of micro-source j capacity, and expression formula is:

$A_{per(i,j)}=\frac{\underset{z\∈L_{i,j}}{\Σ}{P}_{LD,z}}{P_{DG,j}}---\left(1\right)$

Wherein, L _i,jexpress the load aggregation on load i and the shortest communication path of micro-source j, P _{lD, z}for the power of load z on all communication paths, P _{dG, j}for the active power of micro-source j; A _{per (i, j)}numerical value is less, represents that load i is more likely divided to micro-source j;

2.2) load-micro-source ordinal matrix;

Load-micro-source ordinal matrix A _sorit is the sequencing information accounting for this micro-source capacity ratio for recording a type load in power distribution network to all load total amounts on each Wei Yuan UNICOM branch road; Wherein, a type load numbering corresponding row vector, micro-source sequence respective column vector, the then A corresponding to element representation of matrix i-th row jth row _perelements A in matrix _{per (i, j)}in the sequence of the i-th row, sorting arranges from small to large by numerical values recited; Such as A _perelements A in matrix _{per (2,2)}corresponding numerical value is 1.2, after its place row sorts from small to large, is positioned at first, so A _sora corresponding in matrix _{sor (2,2)}the numerical value of element is 1;

2.3) load-micro-source ownership matrix;

Load-micro-source ownership matrix A _belrepresent according to certain rule, in power distribution network, a type load is divided to the information in certain micro-source; Wherein, a type load numbering corresponding row vector, micro-source numbering respective column vector, in matrix, the i-th row, j column element are 1, represent load i and belong to micro-source j, and 0 represents that load i does not belong to micro-source j;

2.4) micro-source-micro-source networking matrix;

When certain micro-source cannot meet the larger type load of one or more capacity, carry out networking between Wei Yuan and micro-source to meet the larger type load of capacity; Networking matrix A _unirow vector and column vector all represent that micro-source is numbered, if the i-th row, jth column element are 1, represent that networking is carried out in an i-th micro-source and jth micro-source, form an island network system; In the algorithm starting stage, each self-forming in each micro-source network, therefore, networking matrix is unit matrix;

3) initial disaggregation is searched for;

For realizing the division of a type load, needing to formulate relevant rule set and being used for solving of heuristic search algorithm; Propose 2 kinds of heuristic rules for searching for initial disaggregation, rule is described below:

Rule 1: in a final type load division result, a type load i can not belong to A _perin matrix i-th row, numerical value is more than or equal to micro-source of 1; Because, if A _perin matrix, an i-th row jth element is more than or equal to 1, shows that load i is greater than micro-source j capacity to the load total amount on the j communication path of micro-source, does not meet constraint condition (2), i.e. formula (6);

Rule 2: compared to A _sorin the larger micro-source j of the i-th line ordering sequence number, a type load i more easily belong to the less micro-source j of sequence sequence number ^*; Because sequence sequence number is little, show that a type load i is to micro-source j ^*it is less that load total amount on communication path accounts for micro-source capacity ratio, and this micro-source can hold a more type load;

According to above-mentioned heuristic rule, the search step of initial network disaggregation is described below: first, according to topology of networks after distribution network failure, draws load-micro-source scaling matrices A _per, load-micro-source ordinal matrix A _sor; Secondly, according to regular 1 Sum fanction 2, draw load-micro-source ownership matrix A _bel; Finally, according to matrix A _bel, the initial division result of a type load after intelligent distribution network fault can be drawn;

4) power verification;

For realizing the power verification of each island network, following heuristic rule is proposed:

Rule 3: the island network at all Wei Yuan places must meet power-balance verification, namely the capacity in this micro-source must be more than or equal to the power summation of all loads on all type loads of this Wei Yuan place network and a type load to micro-source communication path; Check formula is:

$\underset{z\∈D_j}{\Σ}{P}_z+P_{loss,j}\≤P_{DG,j}---\left(2\right)$

Wherein, P _zrepresent the power of load z, set D _jrepresent that type loads all in the island network at j place, micro-source and all type loads are to all loads on the j communication path of micro-source; P _{loss, j}represent the network loss of isolated island by obtaining after Load flow calculation residing for micro-source j;

Rule 4: if micro-source j does not meet power-balance verification, then reject A corresponding to a type load in micro-source j _perthe load that in matrix, numerical value is maximum, and this load is divided to its A _sormicro-source of the rear cis-position of corresponding row sequence in matrix;

According to above-mentioned heuristic rule, the step of power verification is as described below: for the splitting scheme of a type load initial in step (3), regularly 3 carry out power verification to the network at each Wei Yuan place, all micro-sources all verify one and take turns and be designated as an iteration; If in an iteration, all micro-sources all meet rule 3, then enter step 6), if do not meet, then regularly 4 revise, upgrade matrix A _bel, then carry out next round iteration, until all micro-sources all meet rule 3; If certain load is repeatedly rejected by micro-source, illustrate that single micro-source cannot carry this load, then enter networking correction step, i.e. step 5);

5) networking correction;

If certain load is repeatedly rejected by micro-source, illustrate that single micro-source cannot carry this load, need networking operation be carried out; For realizing the networking correction between each micro-source, following heuristic rule is proposed:

Rule 5: if 2 to be divided in any micro-source by a type load i according to rule, the network at this Wei Yuan place all can be caused not meet formula (2), then by A corresponding to this load _sortwo that sort the most forward in matrix i-th row micro-source composition networkings, this type load is divided in this networking accordingly; If after inspection, still do not meet formula (2), then first three micro-source should form networking, the like;

According to above-mentioned heuristic rule, the step of networking correction is as follows: record the number of times that each load is disallowable in each iterative process, if the number of times disallowable in this iteration of certain load exceedes micro-source sum, then carry out networking correction according to rule 5, simultaneously and upgrade matrix A _per, A _beland A _uni, and be back to power checking procedure and carry out next round iteration, successively each networking is verified;

6) a type load partitioning algorithm convergence inspection;

Judge whether that all micro-sources all meet formula (2), if meet, then enter step (7), if do not meet, get back to step (4), continue to carry out power verification to the network at each Wei Yuan place, carry out next round iteration;

7) a type load division result is exported;

Heuritic approach terminates, and exports a type load division result, obtains the switch sequence number set closed;

8) algorithm transition;

According to closed switch sequence number set, to not determining that the switch-linear hybrid of on off state is the optimized variable of quantum discrete particle cluster algorithm;

9) quanta particle swarm optimization optimum configurations;

The dimension of quantum discrete particle cluster algorithm, iterations and corresponding parameter value are set;

10) initialization of quanta particle;

The positional value x of each particle of initialization ^kelite's collection of (i.e. the quantity of state of switch), quantum bit position, rotation angle, local optimum vector x p and non-domination solution, picks out non-domination solution that elite concentrates crowding distance minimum as global optimum vector x g;

11) objective function calculates;

According to the positional value of each particle, draw the running status of power distribution network in conjunction with a type load division result, and according to objective function (3) and (4), calculate the adaptive value of each particle;

12) parameter upgrades;

According to the more new formula of quantum particle swarm, upgrade the bit of quantum rotation angle guiding value, quantum rotation angle and quanta particle successively;

13) positional value and optimal vector upgrade;

According to the more new formula of quantum particle swarm, upgrade positional value x ^k, and local optimal vector;

14) non-domination solution screening;

According to the target function value of each particle, filter out Pareto optimum solution, and it is collected stored in elite;

15) elite collects screening;

Current population is carried out to the computing of non-domination solution, the non-domination solution finally obtained is put into elite's collection; Select non-domination solution that elite concentrates the spacing of individuality maximum as globally optimal solution;

16) computing is eliminated;

Utilize and improve microhabitat filtering technique, carry out superseded computing to the non-domination solution that elite concentrates, elite concentrates the diversity of population;

17) quanta particle swarm optimization test for convergence;

Whether check algorithm restrains; If so, then step 18 is entered); If not, then get back to step 11);

18) result exports;

According to the net result that heuritic approach and quanta particle swarm optimization draw, the fail-over policy of output distribution net.