CN104112165A - Intelligent power distribution network fault recovery method based on multi-target discrete particle swarm - Google Patents

Abstract

The invention discloses an intelligent power distribution network fault recovery method based on a multi-target discrete particle swarm. The method comprises the following steps: 1), initializing parameters; 2), initializing a position and a speed of a discrete particle swarm optimization algorithm, and according to intelligent power distribution network island dividing and load recovery algorithms, calculating each fitness value fitness k relative to a multi-target function for each particle; 3), based on a fitness control concept, performing classification of a control population and a non-control population; 4), according to a corresponding updating rule, updating a particle position and a particle speed in the control population; 5), performing dynamic exchange between the control population and the non-control population; 6), detecting whether a maximum iteration frequency is reached, and if the maximum iteration frequency is reached, skipping to step (4), and otherwise, entering step 7); and 7), outputting a final optimization result.

Description

Intelligent distribution network fault recovery method based on multiple goal discrete particle cluster

Technical field

The present invention relates to a kind of containing distributed power generation distribution fault restoration methods, particularly a kind of intelligent distribution network fault recovery method based on multiple goal discrete particle cluster.

Background technology

Intelligent distribution network is the direction of following Electric Power Network Planning and construction." intelligent grid " refers to modern electric power supply system, and it can be monitored, the operation of protection and its interconnected element of Automatic Optimal, and then effectively meets user security, reliable and multifarious power demands.The intellectuality of distribution and electricity consumption is the emphasis of intelligent grid research, and following intelligentized power distribution network is by the chief component that is intelligent grid.Intelligentized power distribution network requires from the distribution of traditional passive type to active transformation, and this distribution is conducive to access distributed power generation unit, realizes the intellectuality operation that supply side and user's side can participate in real time.This novel network structure of micro-electrical network is just to realize a kind of effective mode of active power distribution network, develops and the concept of extending micro-electrical network can promote the extensive access of distributed power source and regenerative resource to make traditional electrical network to the transition of intelligent network.

Distribution network failure recovers after the generation of assignment electric network fault, by determining optimum switch combination scheme, realize the targets such as recovery dead electricity load is maximum, switching manipulation least number of times, loss minimization, meet conditions such as recovering rear power distribution network is connective, radial, feeder line nonoverload simultaneously.It is the nonlinear optimal problem of a multiple goal, multiple constraint that distribution network failure recovers, and the solution finally obtaining is a series of Switch State Combination in Power Systems.Traditional distribution network failure recovers method for solving and mainly contains intelligent optimization method two classes such as heuristic search and genetic algorithm, tabu search algorithm, ant group algorithm, many Agent Theory.Heuristic search is converted into corresponding processing rule by expertise, but the original state of system is very large on Search Results impact, and the stability of algorithm is good not.In intelligent algorithm, genetic algorithm is most widely used, to optimization problem, can not lead and continuity requirement, only need a fitness function or performance index, and have global convergence, its major defect is that " Premature Convergence " problem and speed of convergence are difficult to meet the needs of controlling in real time.Particle swarm optimization algorithm is as a kind of novel based on Swarm Intelligent Computation method, demonstrated powerful advantage when the non-linear ill optimization problem such as discontinuous, non-differentiability that solves that classic optimisation algorithm is difficult to solve and combinatorial optimization problem.Compare with other evolution algorithms, it have thought simple, easily realize, the advantage such as adjustable parameter is less and effect is obvious,

Although PSO algorithm has good convergence and precision, still cannot meet micro-distribution network system its computing time to being the requirement of real-time, and arranging of algorithm parameter affects on algorithm performance larger

Summary of the invention

Research for existing system fault recovery, mostly for be traditional distribution network, but the introducing due to all kinds of distributed energies in intelligent distribution network system, make traditional distribution network failure recovery algorithms no longer meet new requirement, the invention provides a kind of method that can adapt to the intelligent distribution network system fault recovery based on discrete particle cluster algorithm novel intelligent distribution network system, that efficiency is higher.Fault recovery method different from the past, need to formulate corresponding fault recovery heuristic rule, and content of the present invention, without corresponding fault recovery rule, only relies on intelligent optimization algorithm to solve the problem of this multi-constraint condition of fault recovery; And project of the present invention can be applicable to the higher intelligent distribution network system of current distributed energy (DG) permeability.The process flow diagram of the fault recovery method of the intelligent distribution system based on multiple goal Discrete Particle Swarm Optimization Algorithm (BPSO) as shown in Figure 1.

Intelligent trouble restoration methods based on multiple goal discrete particle cluster intelligent optimization algorithm, comprises the following steps

1) initiation parameter;

2) position, the speed in initialization Discrete Particle Swarm Optimization Algorithm, and according to the algorithm of the division of intelligent distribution network isolated island and load restoration, each fitness value fitness to each calculating particles with respect to multiple objective function _k;

3) concept based on fitness domination, the classification of propping up mating group and Fei Zhi mating group;

4) according to corresponding update rule, particle position and speed in a mating group are upgraded;

5) prop up the dynamic exchange between mating group and Fei Zhi mating group;

6) detect whether to reach maximum iteration time, if reach maximum iteration time, jump procedure (4), otherwise enter step 7);

7) export final optimum results.

For further setting forth the novelty of content of the present invention, described in more detailed step thes contents are as follows:

Step 1, initiation parameter.

1-1) determine maximum iteration time, population, number of dimensions, the study factor, the value of inertia weight and the computing formula of all kinds of objective functions and the relevant parameter in Discrete Particle Swarm Optimization Algorithm.The inventive method---in the fault recovery method based on discrete particle cluster intelligent distribution network, comprised the objective function of three aspects:

I) recovery dead electricity load as much as possible

$\begin{array}{cc}\max & f_1=\underset{i\∈D}{\mathrm{\Σ}}{\mathrm{\λ}}_i{P}_i\end{array}---\left(1\right)$

Wherein, P _ithe size that represents dead electricity load; λ _ithe weight coefficient that represents dead electricity load i, represents the priority level of loading; D represents the load aggregation of whole system.

Ii) number of operations of less as far as possible switch:

$\begin{array}{cc}\min & f_2=\underset{k\∈T_s}{\mathrm{\Σ}}{K}_k\end{array}---\left(2\right)$

Wherein, K _kthe state that represents switch, 1 represents closure, 0 represents to open; T _srepresent the set of whole switch.

Iii) after recovering, the network loss of whole system is as much as possible little:

$\begin{array}{cc}\min & f_3=\underset{l\∈N_l}{\mathrm{\Σ}}{I}_l^2{R}_l\end{array}---\left(3\right)$

Wherein, f ₃represent to form the active power loss of power supply isolated island, I _lthe electric current that represents branch road, N _lrepresent the power supply branch road in whole system.

In project of the present invention, the constraint condition of fault recovery mainly comprises

1-2) within the scope of the feasible zone of constraint condition,

Step 2, the position of discrete particle and speed initialization.

2-1) the positional value of initialization discrete particle

In project of the present invention, the positional value of each particle is the state value of all kinds of switches, and 1 represents closure, and 0 represents to open.Therefore,, in the scope of constraint condition, define at random the positional value (be random cut-off each switch to form corresponding isolated island) of all kinds of particles.Constraint condition in project of the present invention mainly comprises:

I) voltage of node constraint, the voltage of each node should keep in allowed limits, within the maximal value and minimum value of voltage adaptable.

Ii) power constraint of branch road, the power on each branch road should not surpass the permission power of its maximum.

Iii) not dead electricity constraint of a type load, after fault in formed each isolated island, load level is that the load of a class must guarantee not dead electricity.

Vi) capacity-constrained of isolated island, the total load in each isolated island and total losses sum can not surpass DG generating total amounts all in this isolated island.

2-2) the speed of initialization discrete particle cluster.

At random discrete particle cluster is carried out to the initialization of speed, and the initial value of speed is limited between 0 to 1.

2-3) calculate the adaptive value of each particle.

The positional value (being closure or the open mode of switch) obtaining after random according to all kinds of particles, is input in whole intelligent distribution network system, forms corresponding isolated island, and carries out trend calculating to meeting the isolated island of constraint condition, draws corresponding via net loss.And the result obtaining according to these to calculate be the value of objective function (1)-(3), draw the fitness value fitness of each particle _k.

Step 3, the classification of population.

Concept based on fitness domination, is divided into two subgroups by whole population, is respectively non-domination SUBGROUP P _ set and domination subgroup NP_set.Wherein in P_set subgroup, the number of particle is n ₁, in NP_set, the number of particle is n ₂.

Step 4, the renewal of particle position and speed.

4-1) according to the velocity amplitude renewal of formula to particle below:

${v_k}^{t+1}={v_k}^t+c_1(p_{\mathrm{best}}-x_k^t)=c_2(g_{\mathrm{best}}-x_k^t)---\left(4\right)$

Wherein, x _kthe positional value that represents particle, v _k ^trepresent the velocity amplitude of particle k when the t time iterative computation, p _bestthe historical optimal value that represents each particle.G _bestfor the optimal value of selecting in non-domination subgroup at random.

4-2) according to formula below, the positional value of particle is upgraded:

${x_k}^{t+1}=\left\{\begin{array}{c}1,\mathrm{if\; rand}\left(\right)<\mathrm{sig}\left({v_k}^{t+1}\right)\\ 0,\mathrm{else}\end{array}\right.---\left(5\right)$

Wherein,

$\mathrm{sig}\left({v_k}^{t+1}\right)=\frac{1}{1+\exp (-v_k^{t+1})}---\left(6\right)$

Step 5, the dynamic exchange between a mating group and Fei Zhi mating group.

Each particle in NP_set subgroup and each particle in P_set subgroup are compared one by one, and carry out corresponding swap operation, so that all particles in final NP_set subgroup are domination particle, i.e. the adaptive value adaptive value of the particle in P_set subgroup that is all dominant.And, finally delete the particle repeating in P_set subgroup and NP_set subgroup.

Step 6, the test for convergence of method.

Check whole Discrete Particle Swarm Optimization Algorithm whether to reach maximum iterations, if reach iterations, reached the maximum times setting, jump out iterative step, enter next step (7); If also do not reach maximum iteration time, get back to step (4).

Step 7, exports final optimum results.

According to final resulting result, it is the state value (on off state value) of particle, draw final isolated island scheme and load restoration scheme, and export final three-dimensional Pareto optimality curved surface, for operator, select corresponding switching manipulation amount and load restoration scheme.

Advantage of the present invention is: when intelligent distribution network is carried out to fault recovery, not needing to formulate corresponding fault recovery criterion divides to carry out artificial fault, only need to know the structural parameters of network, all kinds of load level and can, for the number of switches of controlling, utilize intelligent optimization algorithm-discrete particle cluster algorithm to carry out solving of fault recovery scheme.And, operating personnel can according in final resulting Pareto optimality curved surface and actual environment to the choice of all kinds of objective functions and deflection, carry out the selection of optimal case.The inventive method is utilized up-to-date intelligent optimization algorithm, can obtain faster the optimal case of fault recovery, can better meet the true-time operation demand in actual intelligent distribution network.

Accompanying drawing explanation

Fig. 1 is process flow diagram of the present invention

Fig. 2 is applicable to the intelligent distribution network structural drawing of this law method

The resulting final isolated island division result of Fig. 3 the inventive method

The DG parameter of Fig. 4 intelligent distribution network

The line parameter circuit value of Fig. 5 intelligent distribution network

The load parameter of Fig. 6 intelligent distribution network

The load restoration result that Fig. 7 is final

Embodiment

With reference to accompanying drawing

Intelligent trouble restoration methods based on multiple goal discrete particle cluster intelligent optimization algorithm, comprises the steps:

Step 1, initiation parameter.

1-1) determine maximum iteration time, population, number of dimensions, the study factor, the value of inertia weight and the computing formula of all kinds of objective functions and the relevant parameter in Discrete Particle Swarm Optimization Algorithm.The inventive method---in the fault recovery method based on discrete particle cluster intelligent distribution network, comprised the objective function of three aspects:

I) recovery dead electricity load as much as possible

$\begin{array}{cc}\max & f_1=\underset{i\&Element;D}{\mathrm{\&Sigma;}}{\mathrm{\&lambda;}}_i{P}_i\end{array}---\left(1\right)$

Wherein, P _ithe size that represents dead electricity load; λ _ithe weight coefficient that represents dead electricity load i, represents the priority level of loading; D represents the load aggregation of whole system.

Ii) number of operations of less as far as possible switch:

$\begin{array}{cc}\min & f_2=\underset{k\&Element;T_s}{\mathrm{\&Sigma;}}{K}_k\end{array}---\left(2\right)$

Wherein, K _kthe state that represents switch, 1 represents closure, 0 represents to open; T _srepresent the set of whole switch.

Iii) after recovering, the network loss of whole system is as much as possible little:

$\begin{array}{cc}\min & f_3=\underset{l\&Element;N_l}{\mathrm{\&Sigma;}}{I}_l^2{R}_l\end{array}---\left(3\right)$

Wherein, f ₃represent to form the active power loss of power supply isolated island, I _lthe electric current that represents branch road, N _lrepresent the power supply branch road in whole system.

In project of the present invention, the constraint condition of fault recovery mainly comprises

1-2) within the scope of the feasible zone of constraint condition,

Step 2, the position of discrete particle and speed initialization.

2-1) the positional value of initialization discrete particle

In project of the present invention, the positional value of each particle is the state value of all kinds of switches, and 1 represents closure, and 0 represents to open.Therefore,, in the scope of constraint condition, define at random the positional value (be random cut-off each switch to form corresponding isolated island) of all kinds of particles.Constraint condition in project of the present invention mainly comprises:

I) voltage of node constraint, the voltage of each node should keep in allowed limits, within the maximal value and minimum value of voltage adaptable.

Ii) power constraint of branch road, the power on each branch road should not surpass the permission power of its maximum.

Iii) not dead electricity constraint of a type load, after fault in formed each isolated island, load level is that the load of a class must guarantee not dead electricity.

Vi) capacity-constrained of isolated island, the total load in each isolated island and total losses sum can not surpass DG generating total amounts all in this isolated island.

2-2) the speed of initialization discrete particle cluster.

At random discrete particle cluster is carried out to the initialization of speed, and the initial value of speed is limited between 0 to 1.

2-3) calculate the adaptive value of each particle.

The positional value (being closure or the open mode of switch) obtaining after random according to all kinds of particles, is input in whole intelligent distribution network system, forms corresponding isolated island, and carries out trend calculating to meeting the isolated island of constraint condition, draws corresponding via net loss.And the result obtaining according to these to calculate be the value of objective function (1)-(3), draw the fitness value fitness of each particle _k.

Step 3, the classification of population.

Concept based on fitness domination, is divided into two subgroups by whole population, is respectively non-domination SUBGROUP P _ set and domination subgroup NP_set.Wherein in P_set subgroup, the number of particle is n ₁, in NP_set, the number of particle is n ₂.

Step 4, the renewal of particle position and speed.

4-1) according to the velocity amplitude renewal of formula to particle below:

${v_k}^{t+1}={v_k}^t+c_1(p_{\mathrm{best}}-x_k^t)=c_2(g_{\mathrm{best}}-x_k^t)---\left(4\right)$

Wherein, x _kthe positional value that represents particle, v _k ^trepresent the velocity amplitude of particle k when the t time iterative computation, p _bestthe historical optimal value that represents each particle.G _bestfor the optimal value of selecting in non-domination subgroup at random.

4-2) according to formula below, the positional value of particle is upgraded:

${x_k}^{t+1}=\left\{\begin{array}{c}1,\mathrm{if\; rand}\left(\right)<\mathrm{sig}\left({v_k}^{t+1}\right)\\ 0,\mathrm{else}\end{array}\right.---\left(5\right)$

Wherein,

$\mathrm{sig}\left({v_k}^{t+1}\right)=\frac{1}{1+\exp (-v_k^{t+1})}---\left(6\right)$

Step 5, the dynamic exchange between a mating group and Fei Zhi mating group.

Each particle in NP_set subgroup and each particle in P_set subgroup are compared one by one, and carry out corresponding swap operation, so that all particles in final NP_set subgroup are domination particle, i.e. the adaptive value adaptive value of the particle in P_set subgroup that is all dominant.And, finally delete the particle repeating in P_set subgroup and NP_set subgroup.

Step 6, the test for convergence of method.

Check whole Discrete Particle Swarm Optimization Algorithm whether to reach maximum iterations, if reach iterations, reached the maximum times setting, jump out iterative step, enter next step (7); If also do not reach maximum iteration time, get back to step (4).

Step 7, exports final optimum results.

According to final resulting result, it is the state value (on off state value) of particle, draw final isolated island scheme and load restoration scheme, and export final three-dimensional Pareto optimality curved surface, for operator, select corresponding switching manipulation amount and load restoration scheme.

Advantage of the present invention is: when intelligent distribution network is carried out to fault recovery, not needing to formulate corresponding fault recovery criterion divides to carry out artificial fault, only need to know the structural parameters of network, all kinds of load level and can, for the number of switches of controlling, utilize intelligent optimization algorithm-discrete particle cluster algorithm to carry out solving of fault recovery scheme.And, operating personnel can according in final resulting Pareto optimality curved surface and actual environment to the choice of all kinds of objective functions and deflection, carry out the selection of optimal case.The inventive method is utilized up-to-date intelligent optimization algorithm, can obtain faster the optimal case of fault recovery, can better meet the true-time operation demand in actual intelligent distribution network.

2, case analysis

It is basis that this project be take in U.S. PG & E69 node distribution system, and introduces therein distributed power source, and the structural drawing of original system as shown in Figure 2.The distributed power source parameter of introducing in intelligent distribution network as shown in Figure 4.Line parameter circuit value as shown in Figure 5.The parameter of each type load as shown in Figure 6.Suppose that three-phase ground fault has occurred at circuit 2-3 place, after fault isolation, the system dead electricity of whole fault down stream.

Utilize the inventive method, can obtain the isolated island splitting scheme of whole intelligent distribution network after 2-3 place breaks down, as shown in Figure 3.Final load restoration result as shown in Figure 7.Can find out, the intelligent distribution network fault recovery scheme that the inventive method is drawn, the recovery rate of a type load is 100%, and the network loss of final formed all kinds of isolated islands is also less, can meet the load restoration requirement after intelligent distribution network fault.

Claims (1)

1. the intelligent trouble restoration methods based on multiple goal discrete particle cluster intelligent optimization algorithm, comprises the following steps:

Step 1, initiation parameter.

1.1 determine maximum iteration time, population, number of dimensions, the study factor, the value of inertia weight and the computing formula of all kinds of objective functions and the relevant parameter in Discrete Particle Swarm Optimization Algorithm, the objective function that has comprised three aspects:

1.1.1 recovery dead electricity as much as possible is loaded

$\max f_1=\underset{i\&Element;D}{\mathrm{\&Sigma;}}{\mathrm{\&lambda;}}_i{P}_i---\left(1\right)$

Wherein, P _ithe size that represents dead electricity load; λ _ithe weight coefficient that represents dead electricity load i, represents the priority level of loading; D represents the load aggregation of whole system;

1.1.2 the number of operations of less as far as possible switch:

$\min f_2=\underset{k\&Element;T_s}{\mathrm{\&Sigma;}}{K}_k---\left(2\right)$

Wherein, K _kthe state that represents switch, 1 represents closure, 0 represents to open; T _srepresent the set of whole switch;

1.1.3, after recovering, the network loss of whole system is as much as possible little:

$\min f_3=\underset{l\&Element;N_l}{\mathrm{\&Sigma;}}{I}_l^2{R}_l---\left(3\right)$

Wherein, f ₃represent to form the active power loss of power supply isolated island, I _lthe electric current that represents branch road, N _lrepresent the power supply branch road in whole system; R _lthe impedance parameter that represents branch road.

Step 2, the position of discrete particle and speed initialization;

The positional value of 2.1 initialization discrete particles

The positional value of each particle is the state value of all kinds of switches, and 1 represents closure, and 0 represents to open; Therefore, in the scope of constraint condition, define at random the positional value of all kinds of particles, cut-off each switch to form corresponding isolated island at random; Constraint condition mainly comprises:

2.1.1 the voltage of node constraint, the voltage of each node should keep in allowed limits, within the maximal value and minimum value of voltage adaptable.

2.1.2 the power constraint of branch road, the power on each branch road should not surpass the permission power of its maximum.

2.1.3 not dead electricity constraint of a type load, after fault in formed each isolated island, load level is that the load of a class must guarantee not dead electricity.

2.1.4 the capacity-constrained of isolated island, the total load in each isolated island and total losses sum can not surpass DG generating total amounts all in this isolated island.

The speed of 2.2 initialization discrete particle clusters;

At random discrete particle cluster is carried out to the initialization of speed, and the initial value of speed is limited between 0 to 1;

2.3 calculate the adaptive value of each particle;

The positional value obtaining after random according to all kinds of particles, the closure of switch or open mode, be input in whole intelligent distribution network system, forms corresponding isolated island, and carry out trend calculating to meeting the isolated island of constraint condition, draws corresponding via net loss.And the result obtaining according to these to calculate be the value of objective function (1)-(3), draw the fitness value fitness of each particle _k;

Step 3, the classification of population;

Concept based on fitness domination, is divided into two subgroups by whole population, is respectively non-domination SUBGROUP P _ set and domination subgroup NP_set; Wherein in P_set subgroup, the number of particle is n ₁, in NP_set, the number of particle is n ₂;

Step 4, the renewal of particle position and speed;

4.1 bases below formula are upgraded the velocity amplitude of particle:

${v_k}^{t+1}={v_k}^t+c_1(p_{\mathrm{best}}-x_k^t)+c_2(g_{\mathrm{best}}-x_k^t)---\left(4\right)$

Wherein, x _kthe positional value that represents particle, v _k ^trepresent the velocity amplitude of particle k when the t time iterative computation, p _bestthe historical optimal value that represents each particle; g _bestfor the optimal value of selecting in non-domination subgroup at random;

4.2 upgrade the positional value of particle according to formula below:

${x_k}^{t+1}=\left\{\begin{array}{c}1,\mathrm{ifrand}\left(\right)<\mathrm{sig}\left({v_k}^{t+1}\right)\\ 0,\mathrm{else}\end{array}\right.---\left(5\right)$

Wherein,

$\mathrm{sig}\left({v_k}^{t+1}\right)=\frac{1}{1+\exp (-v_k^{t+1})}---\left(6\right)$

Step 5, the dynamic exchange between a mating group and Fei Zhi mating group;

Each particle in NP_set subgroup and each particle in P_set subgroup are compared one by one, and carry out corresponding swap operation, so that all particles in final NP_set subgroup are domination particle, i.e. the adaptive value adaptive value of the particle in P_set subgroup that is all dominant; And, finally delete the particle repeating in P_set subgroup and NP_set subgroup;

Step 6, the test for convergence of method;

Check whole Discrete Particle Swarm Optimization Algorithm whether to reach maximum iterations, if reach iterations, reached the maximum times setting, jump out iterative step, enter next step (7); If also do not reach maximum iteration time, get back to step (4);

Step 7, exports final optimum results;

According to final resulting result, it is the state value (on off state value) of particle, draw final isolated island scheme and load restoration scheme, and export final three-dimensional Pareto optimality curved surface, for operator, select corresponding switching manipulation amount and load restoration scheme.