CN106992551A - Photovoltaic inversion controller parameter discrimination method based on fuzzy C-mean algorithm and differential evolution hybrid algorithm - Google Patents

Abstract

The invention discloses the photovoltaic inversion controller parameter discrimination method based on fuzzy C-mean algorithm and differential evolution hybrid algorithm, including：1 sets three phase short circuit fault to disturb in photovoltaic inverter grid-connected point, and the active power output to photovoltaic DC-to-AC converter is sampled；2 recognize whole control parameters of photovoltaic inversion controller, wherein outer voltage proportionality coefficient, outer voltage integral coefficient, the final result that the identification result of current inner loop proportionality coefficient is the parameter by fuzzy C-mean algorithm and differential evolution hybrid algorithm；3 set the disturbance of photovoltaic DC Voltage Reference value mutation, and the active power output to photovoltaic DC-to-AC converter again is sampled；4 only recognize the current inner loop integral coefficient of photovoltaic inversion controller, filter inductance parameter by fuzzy C-mean algorithm and differential evolution hybrid algorithm.The present invention can quickly and accurately obtain photovoltaic inversion controller parameter value, so that the emulation dynamic response curve of photovoltaic DC-to-AC converter controller is consistent with actual measurement dynamic response curve.

Description

Photovoltaic inversion controller parameter based on fuzzy C-mean algorithm and differential evolution hybrid algorithm Discrimination method

Technical field

The present invention relates to Power System Analysis technical field, more specifically the present invention relates to a kind of photovoltaic inversion controller Parameter identification method.

Background technology

In recent years, as energy crisis and problem of environmental pollution are increasingly serious, photovoltaic generation is with its cleaning, reproducible spy Put and be increasingly subject to pay attention to, be in the fast-developing and large-scale application stage.Photovoltaic DC-to-AC converter is to realize that photovoltaic system is grid-connected Core apparatus, the grid-connected transient characterisitics to photovoltaic system play decisive role, and photovoltaic inversion controller ginseng is obtained exactly It is several that significance is respectively provided with to grid-connected stability analysis, error protection etc..Current photovoltaic inversion controller parameter is mostly Given by producer or used empirical value, the deviation of model parameter set-point or empirical value and actual value will directly affect model The confidence level of applicability and simulation result, it is therefore necessary to which the higher inverter controller of precision is obtained by the method for parameter identification Model, and verify its applicability.

The parameter identification method of current photovoltaic inversion controller can be divided into two classes：The first kind is opening up for known photovoltaic DC-to-AC converter Flutter the control strategy of structure and inverter controller, but the unknown parameters in model, model is obtained by way of parameter identification Parameter, feature is that the precision of identification model is higher, explicit physical meaning, but needs known models topological structure and control plan Slightly；Photovoltaic DC-to-AC converter is considered as complete black box problem by Equations of The Second Kind, it is only necessary to gather voltage, the electricity of inverter input-output both sides Flow data, Nonlinear Systems Identification modeling is carried out using models such as Wiener or NARX, and gained model can reflect system non-thread Property characteristic, but the precision of identification model is relatively low, and the information such as topological structure and control strategy are lost, it is changed into pure mathematics mould Type, physical significance is not obvious.The target of photovoltaic inversion controller parameter identification is to find one group of parameter to be identified so that based on this The disturbed trajectory of model emulation of parameter is as consistent as possible with measured curve, is substantially optimization problem, can pass through least square The legacy system discrimination methods such as method, maximum-likelihood method, convolution identification method are solved, and can also pass through genetic algorithm, particle cluster algorithm, god Solved through the modern system such as network technique discrimination method.Differential evolution algorithm is because its algorithm is simple, strong robustness and powerful Ability of searching optimum has been widely applied to the parameter identification field of complication system, but the algorithm one significantly has the disadvantage Iteration later stage convergence of algorithm speed is slower.

The content of the invention

The present invention is that there is provided entered based on fuzzy C-mean algorithm and difference to avoid the weak point present in above-mentioned prior art Change the photovoltaic inversion controller parameter discrimination method of hybrid algorithm, to can rapidly and accurately obtain photovoltaic inversion controller mould Shape parameter, so that the emulation dynamic response curve of photovoltaic DC-to-AC converter controller is consistent with actual measurement dynamic response curve.

In order to achieve the above object, the technical solution adopted in the present invention is：

A kind of photovoltaic inversion controller parameter discrimination method based on fuzzy C-mean algorithm and differential evolution hybrid algorithm of the present invention The characteristics of be to be recognized as follows：

Step 1: after the grid entry point of photovoltaic DC-to-AC converter sets three phase short circuit fault to disturb, to the photovoltaic DC-to-AC converter Active power exports P_gSampled according to time spacing such as sampling time section progress, and K point of sampling, so as to obtain actual measurement sampling sequence Row are designated as P_g-real={ P_g-real,1,P_g-real,2,...,P_g-real,k,...,P_g-real,K}；P_g-real,kRepresent the photovoltaic DC-to-AC converter K-th actual measurement sampled point active power output P_g；The sampling time section includes：It is one section of steady-state process before disturbance, whole Individual perturbation process and disturbance terminate after one section of steady-state process, 1≤k≤K；

Step 2: according to the actual measurement sample sequence P_g-real, using fuzzy C-mean algorithm and differential evolution hybrid algorithm to light 5 control parameters of volt inverter controller are recognized, including：Outer voltage proportionality coefficient k_pu, outer voltage integral coefficient k_iu, current inner loop proportionality coefficient k_pi, current inner loop integral coefficient k_ii, filter inductance L_s；And be made up of 5 control parameters One control parameter vector [k_pu,k_iu,k_pi,k_ii,L_s], wherein, with the outer voltage proportionality coefficient k_pu, outer voltage integration Coefficient k_iu, current inner loop proportionality coefficient k_piIdentification result be final result：

Step 1, generation initial population P₀, the initial population P₀For the matrix of a M rows N row, it is referred to as one by one per a line Body, M is population scale, and N is the number of parameter to be identified, and N=5；

Utilize formula (1) generation initial population P₀Xth row P₀ ^x, so as to generate the N row of initial population；M∈[5N,10N]；1 ≤x≤N；

P₀ ^x=k_x ^L×ones(M,1)+(k_x ^U-k_x ^L)×rand_(0,1)(M,1) (1)

In formula (1), k_x ^LFor the value lower bound of x-th of control parameter in control parameter vector；k_x ^UFor the control ginseng The value upper bound of x-th of control parameter in number vector；Ones (M, 1) is all 1's matrix that M rows 1 are arranged；rand_(0,1)(M, 1) is M rows 1 The random matrix of row, and all elements are the random number between (0,1) in matrix；

Step 2, set algorithm parameter, including：Population maximum evolutionary generation is G, it is allowed to which error is tolerance, Fuzzy C The action period of means clustering algorithm is T, and mutagenic factor is F, and crossover probability is CR；

It is g to define Evolution of Population algebraically, and initializes the Evolution of Population algebraically g=1；

Step 3, current population P updated by differential evolution algorithm_g-1, the population after being updated is P_g；

Whether step 4, the value for judging g remainders T are zero, if zero, then step 5 is performed, if not zero, then perform step 6；

Step 5, by Fuzzy C-Means Cluster Algorithm come Population Regeneration P_g；

Step 6, calculating population P_gIn all M individual corresponding target function values, wherein object function minimum value note For J_g,min, judge J_g,minWhether ＜ tolerance set up, and show there is the individual for meeting required precision in population if setting up, and Step 8 is performed, otherwise, step 7 is performed；

Step 7, judge whether g=G sets up, show that population has evolved to highest generation if setting up, and perform step 8 to perform, Otherwise, g+1 is assigned to after g, return to step 3 is performed；

Step 8, population P_gMiddle object function minimum value J_g,minCorresponding individual as recognizes obtained photovoltaic inversion control Control parameter vector [the k of device_pu,k_iu,k_pi,k_ii,L_s]；

Step 3: setting direct voltage reference value u in the photovoltaic inversion controller_dc-refMutation disturbance, and according to step One sample mode exports P to the active power of photovoltaic DC-to-AC converter_gSampled as measured data；

Step 4: keeping outer voltage proportionality coefficient k in step 2_pu, outer voltage integral coefficient k_iu, current inner loop ratio Example coefficient k_piIdentification result it is constant, further according to the measured data, utilize the fuzzy C-mean algorithm and differential evolution hybrid algorithm Only to current inner loop integral coefficient k_iiWith filter inductance L_sRecognized, so as to obtain the control parameter vector [k_pu,k_iu, k_pi,k_ii,L_s]。

Photovoltaic inversion controller parameter identification side of the present invention based on fuzzy C-mean algorithm and differential evolution hybrid algorithm The characteristics of method, lies also in, and in the step 3 is to carry out population P according to the following procedure based on differential evolution algorithm_g-1Update：

Step I, mutation operation

Using formula (2) to current population P_g-1S-th of individual P_g-1,sCarry out s-th after mutation operation is made a variation Body H_g-1,s, so as to current population P_g-1In M individual carry out mutation operation, M after make a variation is individual, so as to constitute G-1 is for Variation Matrix H_g-1=[H_g-1,1,H_g-1,2,...,H_g-1,s,...,H_g-1,M]^T, wherein 1≤p1≤M, 1≤p2≤M, 1≤ P3≤M, 1≤s≤M, and p1 ≠ p2 ≠ p3 ≠ s；

H_g-1,s=P_g-1,p1+(P_g-1,p2-P_g-1,p3)×F (2)

In formula (2), P_g-1,p1、P_g-1,p2、P_g-1,p3For from current population P_g-1In optional three individuals；F is variation The factor, F ∈ (0,2)；

Step II, to g-1 for Variation Matrix H_g-1Xth column element in more bound component be modified, the member of crossing the border Element refers to be less than value lower bound k_x ^LOr more than value upper bound k_x ^UElement, to less than k_x ^LElement be modified to k_x ^L, to more than k_x ^U Element be modified to k_x ^U, so as to complete to g-1 for Variation Matrix H_g-1N column elements in more bound component amendment；

Step III, crossover operation

The u parameter Vs of the g-1 for s-th of individual of cross matrix is obtained using formula (3)_g-1,s,u, so as to obtain g- N number of parameter of s-th of individual of 1 generation cross matrix, and then obtain N number of parameters of the g-1 for M individual of cross matrix；1≤ u≤N；

In formula (3), P_g-1,s,uFor current population P_g-1S-th individual u-th of parameter；H_g-1,s,uMade a variation for g-1 generations Matrix H_g-1S-th individual u-th of parameter；CR is crossover probability, CR ∈ (0,1)；

Step IV, g-1 obtained for cross matrix V by formula (4)_g-1S-th of individual corresponding target function value J_v(g-1),s, so as to obtain g-1 for cross matrix V_g-1M individual corresponding target function values；

In formula (4), P_{g-sim,k,v(g-1),s}For the control parameter vector [k of photovoltaic inversion controller_pu,k_iu,k_pi,k_ii,L_s] Be worth is g-1 for cross matrix V_g-1S-th of individual when, the active power output of k-th of photovoltaic DC-to-AC converter emulation sampled point P_g；

Step V, current population P obtained by formula (5)_g-1S-th of individual corresponding target function value J_p(g-1),s, so that Obtain current population P_g-1M individual corresponding target function values；

In formula (5), P_{g-sim,k,p(g-1),s}For the control parameter vector [k of photovoltaic inversion controller_pu,k_iu,k_pi,k_ii,L_s] It is worth for current population P_g-1S-th of individual when, the active power output P of k-th of photovoltaic DC-to-AC converter emulation sampled point_g；

Step VI, selection operation

G is obtained for population P using formula (6)_gS-th of individual P_g,s, so as to obtain g for population P_gM individual：

In formula (6), V_g-1,sIt is g-1 for cross matrix V_g-1S-th individual；P_g-1,sFor current population P_g-1S-th Individual.

It is to carry out population P according to the following procedure that Fuzzy C-Means Cluster Algorithm is based in the step 5_gUpdate：

Step i, clusters number c is randomly generated,

Step ii, from population P_gMiddle c individual of random selection is designated as v as initial cluster center₁ ^(g),...,v_i ^(g),...,v_c ^(g)；v_i ^(g)Represent i-th of initial cluster center, 1≤i≤c；

Step iii, formula (7) is utilized to obtain subordinated-degree matrix U_c×M ^(g)In the i-th row jth column element μ_ij ^(g), so as to be subordinate to Category degree matrix U_c×M ^(g)；1≤j≤M；

In formula (7), m is fuzzy coefficient, and m ∈ [1 ,+∞), usual m takes 2；||P_g,j-v_i ^(g)||₂For population P_gIn j-th Body P_g,jWith i-th of initial cluster center v_i ^(g)Between Euclidean distance；v_α ^(g)Represent the α initial cluster center, 1≤α≤c；

Step iv, utilize formula (8) calculate i-th of new cluster centreSo as to obtain c new cluster centres：

Step v, the c new cluster centres are defined as to g generation selection set A_g；

Step vi, from population P_gC Different Individual of middle random selection, is defined as g and replaces changing set B_g；

Step vii, from set A_g∪B_gIn find out c minimum individual of target function value, being defined as g generations updates set C_g；

Step viii, pass through formula (9) Population Regeneration P_g：

P_g=(P_g\B_g)∪C_g (9)

Formula (9) represents that g is replaced to change set B_gIn all individuals from population P_gIt is middle to reject, and in g generations, are updated set C_g In all individuals be added to population P_gIn.

Compared with the prior art, beneficial effects of the present invention are embodied in：

1st, the present invention obtains the parameter of photovoltaic inversion controller using fuzzy C-mean algorithm and the identification of differential evolution hybrid algorithm Value, has the advantages that fast convergence rate, identification precision are high, and the model set up based on identification result can react grid entry point exactly Dynamic response characteristic, on analysis photovoltaic generating system it is grid-connected influence have important meaning.

2nd, the present invention is divided by system side three phase short circuit fault and both disturbances of photovoltaic DC Voltage Reference value mutation 5 parameters of step identification photovoltaic inversion controller, make full use of the leading dynamic of system and different disturbances that different perturbation excitations go out Under important parameter, effectively increase the identification precision of parameter.

3rd, a step Fuzzy C-Means Cluster Algorithm is periodically added in differential evolution algorithm by the present invention, it is proposed that mould C averages and differential evolution hybrid algorithm are pasted, conventional differential evolution algorithm is improved, algorithm is enhanced in global search energy Balance between power and local mining ability.

4th, the targeted outer voltage of method proposed by the present invention, current inner loop photovoltaic inversion controller are in direct currents such as energy storage Also very common during source is grid-connected, method has higher expansion and versatility.

Brief description of the drawings

Fig. 1 is photovoltaic DC-to-AC converter involved in the present invention and photovoltaic inversion controller architecture figure；

Fig. 2 is the overview flow chart of photovoltaic inversion controller parameter discrimination method involved in the present invention；

Fig. 3 is the flow chart of fuzzy C-mean algorithm involved in the present invention and differential evolution hybrid algorithm.

Embodiment

In the present embodiment, photovoltaic DC-to-AC converter and photovoltaic inversion controller architecture are as shown in figure 1, photovoltaic DC-to-AC converter is by photovoltaic Inverter controller is controlled, and photovoltaic inversion controller uses common outer voltage, current inner loop control strategy, the control plan Slightly include 5 control parameters altogether, be respectively：Outer voltage proportionality coefficient k_pu, outer voltage integral coefficient k_iu, current inner loop ratio Example coefficient k_pi, current inner loop integral coefficient k_ii, filter inductance L_s；By the reactive current reference value in Fig. 1 photovoltaic inversion controllers It is set to 0, i_qref=0, you can realize that the unity power factor of photovoltaic DC-to-AC converter is grid-connected.

As shown in Fig. 2 the photovoltaic inversion controller parameter discrimination method based on fuzzy C-mean algorithm and differential evolution hybrid algorithm It is to be recognized as follows：

Step 1: after the grid entry point of photovoltaic DC-to-AC converter sets three phase short circuit fault to disturb, to the active of photovoltaic DC-to-AC converter Power output P_gSampled according to time spacing such as sampling time section progress, and K point of sampling, so as to obtain actual measurement sample sequence note For P_g-real={ P_g-real,1,P_g-real,2,...,P_g-real,k,...,P_g-real,K}；P_g-real,kRepresent k-th of photovoltaic DC-to-AC converter Survey the active power output P of sampled point_g；Sampling time section includes：One section of steady-state process, whole perturbation process before disturbance with And disturb one section of steady-state process after terminating, 1≤k≤K；

Parameter identification is dependent on disturbing signal and the selection of observed quantity, active power when being disturbed due to photovoltaic system The general fluctuation for being significantly larger than reactive power of fluctuation, the present invention regard the active power output of photovoltaic DC-to-AC converter as observed quantity.

Step 2: according to actual measurement sample sequence P_g-real, as shown in figure 3, being calculated using fuzzy C-mean algorithm and differential evolution mixing Method is recognized to 5 control parameters of photovoltaic inversion controller, including：Outer voltage proportionality coefficient k_pu, outer voltage integration Coefficient k_iu, current inner loop proportionality coefficient k_pi, current inner loop integral coefficient k_ii, filter inductance L_s；And be made up of 5 control parameters One control parameter vector [k_pu,k_iu,k_pi,k_ii,L_s], wherein, with outer voltage proportionality coefficient k_pu, outer voltage integral coefficient k_iu, current inner loop proportionality coefficient k_piIdentification result be final result：

The purpose of photovoltaic inversion controller parameter identification is to find one group of parameter to be identifiedMake Simulation model dynamic response curve that must be based on the parameter is as consistent as possible with actual measurement dynamic response curve, inverse due to choosing photovoltaic Become the active power output P of device_gFor observed quantity, then object function can be described as：

Therefore, substantially this belongs to an optimization problem, can be solved by optimized algorithm, and the present invention proposes fuzzy C-mean algorithm The problem is solved with differential evolution hybrid algorithm.

Step 1, generation initial population P₀, initial population P₀For the matrix of a M rows N row, it is referred to as an individual, M per a line For population scale, M is the important parameter of differential evolution algorithm, and M is bigger, and the diversity of population is just high, it is not easy to be absorbed in part most It is excellent, but the calculating time can also increase；N is the number of parameter to be identified, and N=5；

Utilize formula (1) generation initial population P₀Xth row P₀ ^x, so as to generate the N row of initial population；M∈[5N,10N]；1 ≤x≤N；

P₀ ^x=k_x ^L×ones(M,1)+(k_x ^U-k_x ^L)×rand_(0,1)(M,1) (1)

In formula (1), k_x ^LFor the value lower bound of x-th of control parameter in control parameter vector；k_x ^UFor in control parameter vector The value upper bound of x-th of control parameter；Ones (M, 1) is all 1's matrix that M rows 1 are arranged；rand_(0,1)(M, 1) is the random of the row of M rows 1 All elements are the random number between (0,1) in matrix, and matrix；

Step 2, set algorithm parameter, including：Population maximum evolutionary generation is G, it is allowed to which error is tolerance, Fuzzy C The action period of means clustering algorithm is T, and mutagenic factor is F, and crossover probability is CR；

It is g to define Evolution of Population algebraically, and initializes Evolution of Population algebraically g=1；

Step 3, current population P updated by differential evolution algorithm_g-1, the population after being updated is P_g；Specifically, base It is to carry out population P according to the following procedure in differential evolution algorithm_g-1Update：

Step I, mutation operation

Using formula (2) to current population P_g-1S-th of individual P_g-1,sCarry out s-th after mutation operation is made a variation Body H_g-1,s, so as to current population P_g-1In M individual carry out mutation operation, M after make a variation is individual, so as to constitute G-1 is for Variation Matrix H_g-1=[H_g-1,1,H_g-1,2,...,H_g-1,s,...,H_g-1,M]^T, wherein 1≤p1≤M, 1≤p2≤M, 1≤ P3≤M, 1≤s≤M, and p1 ≠ p2 ≠ p3 ≠ s；

H_g-1,s=P_g-1,p1+(P_g-1,p2-P_g-1,p3)×F (2)

In formula (2), P_g-1,p1、P_g-1,p2、P_g-1,p3For from current population P_g-1In optional three individuals；F is variation The factor, F ∈ (0,2)；Mutagenic factor is control population diversity and constringent important parameter, when mutagenic factor value is smaller, is planted Constellation variance reduces, and evolutionary process is difficult to jump out local extremum to cause population Premature Convergence；When mutagenic factor is larger, easily jump Go out local extremum, convergence rate can be slack-off；

Step II, to g-1 for Variation Matrix H_g-1Xth column element in more bound component be modified, more bound component is Refer to and be less than value lower bound k_x ^LOr more than value upper bound k_x ^UElement, to less than k_x ^LElement be modified to k_x ^L, to more than k_x ^UMember Element is modified to k_x ^U, so as to complete to g-1 for Variation Matrix H_g-1N column elements in more bound component amendment；

Step III, crossover operation

The u parameter Vs of the g-1 for s-th of individual of cross matrix is obtained using formula (3)_g-1,s,u, so as to obtain g- N number of parameter of s-th of individual of 1 generation cross matrix, and then obtain N number of parameters of the g-1 for M individual of cross matrix；1≤ u≤N；

In formula (3), P_g-1,s,uFor current population P_g-1S-th individual u-th of parameter；H_g-1,s,uMade a variation for g-1 generations Matrix H_g-1S-th individual u-th of parameter；CR is crossover probability, CR ∈ (0,1)；Intersect the factor and can control individual parameter Each dimension is to the degree of participation of intersection, and the intersection bigger convergence rate of the factor is faster, but with the increase for intersecting the factor, convergence is to variation Factor F susceptibility is gradually stepped up.

Step IV, g-1 obtained for cross matrix V by formula (4)_g-1S-th of individual corresponding target function value J_v(g-1),s, so as to obtain g-1 for cross matrix V_g-1M individual corresponding target function values；

In formula (4), P_{g-sim,k,v(g-1),s}For the control parameter vector [k of photovoltaic inversion controller_pu,k_iu,k_pi,k_ii,L_s] Be worth is g-1 for cross matrix V_g-1S-th of individual when, the active power output of k-th of photovoltaic DC-to-AC converter emulation sampled point P_g；

Step V, current population P obtained by formula (5)_g-1S-th of individual corresponding target function value J_p(g-1),s, so that Obtain current population P_g-1M individual corresponding target function values；

In formula (5), P_{g-sim,k,p(g-1),s}For the control parameter vector [k of photovoltaic inversion controller_pu,k_iu,k_pi,k_ii,L_s] It is worth for current population P_g-1S-th of individual when, the active power output P of k-th of photovoltaic DC-to-AC converter emulation sampled point_g；

Step VI, selection operation

G is obtained for population P using formula (6)_gS-th of individual P_g,s, so as to obtain g for population P_gM individual：

In formula (6), V_g-1,sIt is g-1 for cross matrix V_g-1S-th individual；P_g-1,sFor current population P_g-1S-th Individual；

Selection operation uses man-to-man competition surviving policy；

Whether step 4, the value for judging g remainders T are zero, if zero, then step 5 is performed, if not zero, then perform step 6；

Differential evolution algorithm one significantly has the disadvantage that late convergence is slower, by a step Fuzzy C-Means Cluster Algorithm Local Search effect can be played by being added in differential evolution algorithm, thus can accelerate algorithm the convergence speed, but less T values meeting Cause algorithm excessive use species information, cause local convergence, usual T ∈ [5,15] are a rational selections.

Step 5, by Fuzzy C-Means Cluster Algorithm come Population Regeneration P_g；Specifically, based on Fuzzy C-Means Cluster Algorithm It is to carry out population P according to the following procedure_gUpdate：

Step i, clusters number c is randomly generated,

Step ii, from population P_gMiddle c individual of random selection is designated as v as initial cluster center₁ ^(g),...,v_i ^(g),...,v_c ^(g)；v_i ^(g)Represent i-th of initial cluster center, 1≤i≤c；

Step iii, formula (7) is utilized to obtain subordinated-degree matrix U_c×M ^(g)In the i-th row jth column element μ_ij ^(g), so as to be subordinate to Category degree matrix U_c×M ^(g)；1≤j≤M；

In formula (7), m is fuzzy coefficient, and m ∈ [1 ,+∞), usual m takes 2；||P_g,j-v_i ^(g)||₂For population P_gIn j-th Body P_g,jWith i-th of initial cluster center v_i ^(g)Between Euclidean distance；v_α ^(g)Represent the α initial cluster center, 1≤α≤c；

Step iv, utilize formula (8) calculate i-th of new cluster centreSo as to obtain c new cluster centres：

Step v, c new cluster centres are defined as to g generation selection set A_g；

Step vi, from population P_gC Different Individual of middle random selection, is defined as g and replaces changing set B_g；With step ii not Together, now it is necessary for c Different Individual；

Step vii, from set A_g∪B_gIn find out c minimum individual of target function value, being defined as g generations updates set C_g；

Step viii, pass through formula (9) Population Regeneration P_g：

P_g=(P_g\B_g)∪C_g (9)

Formula (9) represents that g is replaced to change set B_gIn all individuals from population P_gIt is middle to reject, and in g generations, are updated set C_g In all individuals be added to population P_gIn；So as to keep Population Size constant；

One step Fuzzy C-Means Cluster Algorithm uses the algorithm generator Population Regeneration P based on colony_g, the algorithm generator Optimization task is divided into 4 independent parts：(1) task is selected：C individual is randomly choosed from population as in initial clustering The heart.(2) task is produced:C new cluster centre is asked for using a step C means clustering algorithms, and is stored in set A.(3) replace Task：C different individual composition set B are randomly choosed from population.(4) more new task：C are selected from set A ∪ B most Excellent individual is used as set C, Population Regeneration P_gFor P_g=(P_g\B)∪C；Essential idea is accelerated by retaining the strategy of elite solution The convergence rate of population；

Step 6, calculating population P_gIn all M individual corresponding target function values, wherein object function minimum value note For J_g,min, judge J_g,minWhether ＜ tolerance set up, and show there is the individual for meeting required precision in population if setting up, and Step 8 is performed, otherwise, step 7 is performed；

Step 7, judge whether g=G sets up, show that population has evolved to highest generation if setting up, and perform step 8 to perform, Otherwise, g+1 is assigned to after g, return to step 3 is performed；

Step 8, population P_gMiddle object function minimum value J_g,minCorresponding individual as recognizes obtained photovoltaic inversion control Control parameter vector [the k of device_pu,k_iu,k_pi,k_ii,L_s]；

Step 3: setting direct voltage reference value u in photovoltaic inversion controller_dc-refMutation disturbance, and according to step one Sample mode exports P to the active power of photovoltaic DC-to-AC converter_gSampled as measured data；

Step 4: keeping outer voltage proportionality coefficient k in step 2_pu, outer voltage integral coefficient k_iu, current inner loop ratio Example coefficient k_piIdentification result it is constant, further according to measured data, using fuzzy C-mean algorithm and differential evolution hybrid algorithm only to electric current Inner ring integral coefficient k_iiWith filter inductance L_sRecognized, so as to obtain control parameter vector [k_pu,k_iu,k_pi,k_ii,L_s]。

In parameter identification, generally according to the complexity of trajectory sensitivity analysis parameter identification, the track of parameter is sensitive Degree is bigger, and the parameter is more easily discernible, for identification photovoltaic inversion controller parameter, and due to choosing, then photovoltaic DC-to-AC converter has Work(power output P_gIt is used as observed quantity, its x-th of control parameter k_xTrace sensitivity be calculated as follows：

After observed quantity is selected, the trace sensitivity of parameter and disturbance are closely related.Verified by calculating with simulation recognition, Draw the following conclusions：Under the disturbance of grid entry point three phase short circuit fault, outer voltage proportionality coefficient k_pu, outer voltage integral coefficient k_iu, current inner loop proportionality coefficient k_piTrace sensitivity and identification precision will far above they in photovoltaic DC voltage reference value Trace sensitivity and identification precision under mutation disturbance, but current inner loop integral coefficient k_iiWith filter inductance L_sTrack it is sensitive Degree and identification precision lack will be less than their trace sensitivities and identification precision under the disturbance of photovoltaic DC Voltage Reference value mutation. Therefore the strategy that the present invention is recognized using a kind of two step：The identified parameters k under the disturbance of grid entry point three phase short circuit fault_pu、k_iu、k_pi, The identified parameters k under the disturbance of photovoltaic DC Voltage Reference value mutation_ii、L_s.The important parameter of system under different disturbances is made full use of, Identification precision is improved with this.

In addition, to current inner loop integral coefficient k_iiWith filter inductance L_sWhen being recognized, algorithm parameter will change, Now the number of parameter to be identified is changed into N=2, and due to M ∈ [5N, 10N], population scale M also will accordingly reduce；

The control parameter value of the photovoltaic inversion controller finally obtained with reference to identification, sets up the emulation of photovoltaic inversion controller Model, compares the dynamic response process of simulation model and realistic model under identical disturbance, to verify that photovoltaic inversion controller is joined The validity of number identification.

Claims (3)

1. a kind of photovoltaic inversion controller parameter discrimination method based on fuzzy C-mean algorithm and differential evolution hybrid algorithm, its feature It is to be recognized as follows：

Step 1: after the grid entry point of photovoltaic DC-to-AC converter sets three phase short circuit fault to disturb, to the active of the photovoltaic DC-to-AC converter Power output P_gSampled according to time spacing such as sampling time section progress, and K point of sampling, so as to obtain actual measurement sample sequence note For P_g-real={ P_g-real,1,P_g-real,2,...,P_g-real,k,...,P_g-real,K}；P_g-real,kRepresent the of the photovoltaic DC-to-AC converter The active power output P of k actual measurement sampled point_g；The sampling time section includes：One section of steady-state process before disturbance, entirely disturb Dynamic process and disturbance terminate after one section of steady-state process, 1≤k≤K；

Step 2: according to the actual measurement sample sequence P_g-real, it is inverse to photovoltaic using fuzzy C-mean algorithm and differential evolution hybrid algorithm 5 control parameters for becoming controller are recognized, including：Outer voltage proportionality coefficient k_pu, outer voltage integral coefficient k_iu, electricity Flow inner ring proportionality coefficient k_pi, current inner loop integral coefficient k_ii, filter inductance L_s；And a control is constituted by 5 control parameters Parameter vector [k processed_pu,k_iu,k_pi,k_ii,L_s], wherein, with the outer voltage proportionality coefficient k_pu, outer voltage integral coefficient k_iu, current inner loop proportionality coefficient k_piIdentification result be final result：

Step 1, generation initial population P₀, the initial population P₀For the matrix of a M rows N row, it is referred to as an individual, M per a line For population scale, N is the number of parameter to be identified, and N=5；

Utilize formula (1) generation initial population P₀Xth row P₀ ^x, so as to generate the N row of initial population；M∈[5N,10N]；1≤x≤ N；

P₀ ^x=k_x ^L×ones(M,1)+(k_x ^U-k_x ^L)×rand_(0,1)(M,1) (1)

In formula (1), k_x ^LFor the value lower bound of x-th of control parameter in control parameter vector；k_x ^UFor the control parameter to The value upper bound of x-th of control parameter in amount；Ones (M, 1) is all 1's matrix that M rows 1 are arranged；rand_(0,1)(M, 1) is what M rows 1 were arranged All elements are the random number between (0,1) in random matrix, and matrix；

Step 2, set algorithm parameter, including：Population maximum evolutionary generation is G, it is allowed to which error is tolerance, fuzzy C-mean algorithm The action period of clustering algorithm is T, and mutagenic factor is F, and crossover probability is CR；

It is g to define Evolution of Population algebraically, and initializes the Evolution of Population algebraically g=1；

Step 3, current population P updated by differential evolution algorithm_g-1, the population after being updated is P_g；

Whether step 4, the value for judging g remainders T are zero, if zero, then step 5 is performed, if not zero, then perform step 6；

Step 5, by Fuzzy C-Means Cluster Algorithm come Population Regeneration P_g；

Step 6, calculating population P_gIn all M individual corresponding target function values, the minimum value of wherein object function is designated as J_g,min, judge J_g,minWhether ＜ tolerance set up, and show there is the individual for meeting required precision in population if setting up, and hold Row step 8, otherwise, performs step 7；

Step 7, judge whether g=G sets up, show that population has evolved to highest generation if setting up, and perform step 8 to perform, otherwise, G+1 is assigned to after g, return to step 3 is performed；

Step 8, population P_gMiddle object function minimum value J_g,minCorresponding individual as recognizes obtained photovoltaic inversion controller Control parameter vector [k_pu,k_iu,k_pi,k_ii,L_s]；

Step 3: setting direct voltage reference value u in the photovoltaic inversion controller_dc-refMutation disturbance, and according to step one Sample mode exports P to the active power of photovoltaic DC-to-AC converter_gSampled as measured data；

Step 4: keeping outer voltage proportionality coefficient k in step 2_pu, outer voltage integral coefficient k_iu, current inner loop proportionality coefficient k_piIdentification result it is constant, further according to the measured data, using the fuzzy C-mean algorithm and differential evolution hybrid algorithm only to electricity Flow inner ring integral coefficient k_iiWith filter inductance L_sRecognized, so as to obtain the control parameter vector [k_pu,k_iu,k_pi,k_ii, L_s]。

2. the photovoltaic inversion controller parameter according to claim 1 based on fuzzy C-mean algorithm and differential evolution hybrid algorithm Discrimination method, it is characterized in that, in the step 3 is to carry out population P according to the following procedure based on differential evolution algorithm_g-1Update：

Step I, mutation operation

Using formula (2) to current population P_g-1S-th of individual P_g-1,sCarry out s-th of individual after mutation operation is made a variation H_g-1,s, so as to current population P_g-1In M individual carry out mutation operation, M after make a variation is individual, so as to constitute the G-1 is for Variation Matrix H_g-1=[H_g-1,1,H_g-1,2,...,H_g-1,s,...,H_g-1,M]^T, wherein 1≤p1≤M, 1≤p2≤M, 1≤p3 ≤ M, 1≤s≤M, and p1 ≠ p2 ≠ p3 ≠ s；

H_g-1,s=P_g-1,p1+(P_g-1,p2-P_g-1,p3)×F (2)

In formula (2), P_g-1,p1、P_g-1,p2、P_g-1,p3For from current population P_g-1In optional three individuals；F is mutagenic factor, F∈(0,2)；

Step II, to g-1 for Variation Matrix H_g-1Xth column element in more bound component be modified, it is described more bound component be Refer to and be less than value lower bound k_x ^LOr more than value upper bound k_x ^UElement, to less than k_x ^LElement be modified to k_x ^L, to more than k_x ^UMember Element is modified to k_x ^U, so as to complete to g-1 for Variation Matrix H_g-1N column elements in more bound component amendment；

Step III, crossover operation

The u parameter Vs of the g-1 for s-th of individual of cross matrix is obtained using formula (3)_g-1,s,u, so as to obtain g-1 generation friendships N number of parameter of s-th of individual of matrix is pitched, and then obtains N number of parameters of the g-1 for M individual of cross matrix；1≤u≤N；

$V_{g-1,s,u}=\left\{\begin{array}{c}{H}_{g-1,s,u},\mathrm{rand}\left(\mathrm{0,1}\right)\≤\mathrm{CR}\\ P_{g-1,s,u},\mathrm{rand}\left(\mathrm{0,1}\right)>\mathrm{CR}\end{array}\right.---\left(3\right)$

In formula (3), P_g-1,s,uFor current population P_g-1S-th individual u-th of parameter；H_g-1,s,uIt is g-1 for Variation Matrix H_g-1S-th individual u-th of parameter；CR is crossover probability, CR ∈ (0,1)；

Step IV, g-1 obtained for cross matrix V by formula (4)_g-1S-th of individual corresponding target function value J_v(g-1),s, So as to obtain g-1 for cross matrix V_g-1M individual corresponding target function values；

$J_{v(g-1),s}=\underset{k=1}{\overset{K}{\Σ}}|P_{g-real,k}-P_{g-sim,k,v(g-1),s}|---\left(4\right)$

In formula (4), P_{g-sim,k,v(g-1),s}For the control parameter vector [k of photovoltaic inversion controller_pu,k_iu,k_pi,k_ii,L_s] value be G-1 is for cross matrix V_g-1S-th of individual when, the active power output P of k-th of photovoltaic DC-to-AC converter emulation sampled point_g；

Step V, current population P obtained by formula (5)_g-1S-th of individual corresponding target function value J_p(g-1),s, so as to obtain Current population P_g-1M individual corresponding target function values；

$J_{p(g-1),s}=\underset{k=1}{\overset{K}{\Σ}}|P_{g-real,k}-P_{g-sim,k,p(g-1),s}|---\left(5\right)$

In formula (5), P_{g-sim,k,p(g-1),s}For the control parameter vector [k of photovoltaic inversion controller_pu,k_iu,k_pi,k_ii,L_s] value be Current population P_g-1S-th of individual when, the active power output P of k-th of photovoltaic DC-to-AC converter emulation sampled point_g；

Step VI, selection operation

G is obtained for population P using formula (6)_gS-th of individual P_g,s, so as to obtain g for population P_gM individual：

$P_{g,s}=\left\{\begin{array}{cc}{V}_{g-1,s},& J_{v\left(g-1\right),s}<J_{p\left(g-1\right),s}\\ P_{g-1,s},& J_{v\left(g-1\right),s}\&GreaterEqual;J_{p\left(g-1\right),s}\end{array}\right.---\left(6\right)$

In formula (6), V_g-1,sIt is g-1 for cross matrix V_g-1S-th individual；P_g-1,sFor current population P_g-1S-th individual.

3. the photovoltaic inversion controller parameter according to claim 1 based on fuzzy C-mean algorithm and differential evolution hybrid algorithm Discrimination method, it is characterized in that, it is to carry out population P according to the following procedure that Fuzzy C-Means Cluster Algorithm is based in the step 5_gUpdate：

Step i, clusters number c is randomly generated,

Step ii, from population P_gMiddle c individual of random selection is designated as v as initial cluster center₁ ^(g),...,v_i ^(g),...,v_c ^(g)；v_i ^(g)Represent i-th of initial cluster center, 1≤i≤c；

Step iii, formula (7) is utilized to obtain subordinated-degree matrix U_c×M ^(g)In the i-th row jth column element μ_ij ^(g), so as to obtain degree of membership square Battle array U_c×M ^(g)；1≤j≤M；

${{\mathrm{\&mu;}}_{ij}}^{\left(g\right)}=\frac{1}{\underset{\mathrm{\&alpha;}=1}{\overset{c}{\&Sigma;}}{\left(\frac{\left|\right|P_{g,j}-{v_i}^{\left(g\right)}|{|}_2}{\left|\right|P_{g,j}-{v_{\mathrm{\&alpha;}}}^{\left(g\right)}|{|}_2}\right)}^{\frac{2}{m-1}}}---\left(7\right)$

In formula (7), m is fuzzy coefficient, and m ∈ [1 ,+∞), usual m takes 2；||P_g,j-v_i ^(g)||₂For population P_gIn j-th individual P_g,jWith i-th of initial cluster center v_i ^(g)Between Euclidean distance；v_α ^(g)Represent the α initial cluster center, 1≤α≤c；

Step iv, utilize formula (8) calculate i-th of new cluster centreSo as to obtain c new cluster centres：

$v_i^{\&prime;\left(g\right)}=\frac{\underset{j=1}{\overset{M}{\&Sigma;}}{\left({{\mathrm{\&mu;}}_{ij}}^{\left(g\right)}\right)}^m{P}_{g,j}}{\underset{j=1}{\overset{M}{\&Sigma;}}{\left({{\mathrm{\&mu;}}_{ij}}^{\left(g\right)}\right)}^m}---\left(8\right)$

Step v, the c new cluster centres are defined as to g generation selection set A_g；

Step vi, from population P_gC Different Individual of middle random selection, is defined as g and replaces changing set B_g；

Step vii, from set A_g∪B_gIn find out c minimum individual of target function value, be defined as g generation renewal set C_g；

Step viii, pass through formula (9) Population Regeneration P_g：

P_g=(P_g\B_g)∪C_g (9)

Formula (9) represents that g is replaced to change set B_gIn all individuals from population P_gIt is middle to reject, and in g generations, are updated set C_gIn All individuals are added to population P_gIn.