CN103258131A - Power circuit component optimization method based on orthogonal learning particle swarm - Google Patents

Abstract

The invention discloses a power circuit component optimization method based on an orthogonal learning particle swarm, and belongs to the power electronic technology and the field of computational intelligence. An orthogonal learning particle swarm optimization with a mutation strategy is used for carrying out optimization on an optimal component design of a power electronic circuit. Firstly, a method of generating a new optimal learning object based on an orthogonal combination mode is designed, and is used for mining information of a historical optimal solution of a particle individual and information of a globally-optimal solution of a swarm in the orthogonal learning particle swarm optimization, and combining a learning object which can guide particles to develop in a better direction, secondly, a mutation operator which can improve diversity of the orthogonal learning particle swarm optimization is designed, and the defect that the orthogonal learning particle swarm optimization easily falls into local optimum is overcome. All components of the power electronic circuit serve as variables needing to be optimized and are coded into individuals of the orthogonal learning particle swarm optimization, optimization is carried out on values of the components of the power electronic circuit through specific optimization processes such as update of the speed, update of the location, mutation operation and update of the optimal learning object of the orthogonal learning particle swarm optimization, and the power circuit component optimization method based on the orthogonal learning particle swarm has important application value in the existing large-scale circuit design and optimization field.

Description

Power circuit element optimization method based on quadrature study population

Technical field:

The present invention relates to power electronic circuit and computational intelligence two big fields, be specifically related to a kind of power circuit element optimization method based on quadrature study particle cluster algorithm.

Technical background:

Since nineteen fifty, power semiconductor device occurred, power electronic circuit had obtained development rapidly and has become a important techniques in the various fields such as industry, commerce, dwelling house, Aero-Space, military affairs and government utility.The model of power electronic circuit, design and analysis are basis and important fields of research in the electronic technology.Power electronic circuit is made up of many element under normal conditions, such as resistor, capacitor and inductor.These electric elements all need could guarantee that through optimal design final circuit can obtain better circuit performance.Along with being on the increase of power electronic circuit scale, suitable element design and parameter are regulated and the slip-stick artist have been formed a great challenge.Therefore, the element of power circuit is optimized to for a kind of important research and uses problem.Aspect power circuit unit piece optimization, traditional method has comprised that the state space method of average, electric current inject equivalent circuit method, sampled data modeling method and state plane analytical approach etc.But these methods are only applicable to specific circuit and need the user to grasp enough circuit design knowledge.In addition, these methods generally all are based on small-signal model, so circuit designers finds to use the circuit reaction under these methods prediction large-signal environment very difficult sometimes.

Because classic method can't satisfy the demand of the power circuit model development that scale increases day by day, since 1970, various optimisation techniques, such as heuristic, the technology such as the method for (or climbing the mountain) and simulated annealing that descend based on knowledge method, gradient are proposed and in succession in order to realize the robotization optimal design of Analog Circuit Design.But these methods are all relatively responsive to initial solution, therefore may fall into local optimum easily in actual use.In recent years, because the application that the evolutionary computation method is achieved success in numerous actual application problem has also occurred using methods such as genetic algorithm, ant group algorithm and particle cluster algorithm to optimize the technical scheme of power electronic circuit design.But, in actual applications, still there are a lot of inferior positions in genetic algorithm and ant group algorithm, for example need to consume a large amount of calculating and just can find more excellent circuit component value relatively, and traditional particle cluster algorithm often can not search out globally optimal solution owing to easy precocity.

In the present invention, in order to keep particle cluster algorithm convergence capabilities fast, avoid algorithm to fall into local optimum simultaneously, designed a kind of particle cluster algorithm based on the quadrature learning strategy, and be applied in the element optimal design of power circuit.The quadrature study particle cluster algorithm of the present invention's design, by the entrained information of particle learning object in the particle cluster algorithm (individual historical optimum and population global optimum just) is carried out tap/dip deep into, the mode of use orthogonal experiment makes up the learning object an of the best, the flight of guiding particle finally makes algorithm can possess stronger ability of searching optimum.In quadrature study particle cluster algorithm, when using the orthogonal experiment mode to make up best learning object, the key element correspondence of orthogonal test be the dimension of waiting the problem of finding the solution, and the varying level correspondence of each key element is two kinds of different selections: select individual historical optimum or the global optimum of population.Because the orthogonal experiment mode can comprehensively be searched for interblock space by less different key elements and the combination of varying level, therefore can be by individual historical orthogonal experiment optimum and population global optimum being combined into the learning object an of the best.On the other hand, the quadrature study particle cluster algorithm of the present invention's design also has been equipped with a mutation operation: after having finished the renewal of speed and position under the guiding of particle at best learning object, new position will select some dimensions to make a variation according to certain probability, strengthen the diversity of algorithm, be conducive to improve the ability of searching optimum of algorithm.

Summary of the invention:

The present invention is with a kind of element optimization problem that is used for power circuit with the quadrature study particle cluster algorithm of mutation operation, the advantage of this technology is: 1, based on particle swarm optimization algorithm, this technology has the characteristics of quick convergence, can find the optimum element value of power circuit very soon; 2, owing to used quadrature learning strategy and mutation operation, this technology can overcome the defective that falls into local optimum easily.Therefore technology of the present invention can the realization of fast and stable ground to the optimization of power electronic circuit element value.

The concrete step of using the tactful quadrature study particle cluster algorithm of band variation to find the solution power electronic circuit element optimization problem is described below:

The parameter of step 1) initialization quadrature study particle cluster algorithm comprises the inertia weights omega is made as 0.5, and study factor c is made as 2.0, and population scale N is made as 40, and maximum evolutionary generation G is made as 500; Generate N particle at random and form population, the position of each particle i and velocity encoded cine are expressed as X respectively _i=[x _I1, x _I2..., x _ID] and V _i=[v _I1, v _I2..., v _ID]; Wherein D is code length, and is identical with the number of elements of circuit; Position X _iIn each dimension value representation should the dimension circuit component value; The locative situation of change of speed; Assess the adaptive value of all particles, make the historical optimal location P of particle i _i=[p _I1, p _I2..., p _ID] be current location X _i, the position G=[g of global optimum of whole population ₁, g ₂..., g _D] be all historical optimal location P _iIn best that, the best learning object O of each particle i is set simultaneously _i=[o _I1, o _I2..., o _ID] be current individual historical optimal location P _iAnd best learning object inefficacy algebraically t _i=0.

Step 2) to each particle i, by its best learning object O _iTo speed V _iUpgrade; For V _iEach the dimension v _Id(1≤d≤D), more new formula is accordingly: v _Id=ω * v _Id+ c * r * (o _Id-x _Id); Wherein r is the random number between interval [0,1].

Step 3) is to each particle i, and its current location is upgraded in use location more new formula, for X _iEach the dimension x _Id(1≤d≤D) accordingly more new formula be: x _Id=x _Id+ v _Id

Step 4) is used the position X after the variation strategy upgrades particle i _iMake a variation, strengthen the diversity of algorithm.Concrete method is: generate a random number between [0,1] at random earlier, if the value that this random number less than variation Probability p m (pm=0.01), selects certain one dimension of particle also will tie up is being set in optional scope at random.

Step 5) is upgraded position X after the variation to particle i _iCarry out the assessment of adaptive value, if new fitness function value is than its historical optimal location P _iThe fitness function value better, then with P _iBe set to X _i, and with the best learning object inefficacy algebraically t of particle i _iBe made as 0, otherwise with the best learning object inefficacy algebraically t of particle i _iBe made as t _i+ 1; In addition, judge new P _iWhether more excellent than the position G of global optimum of population, if then G is replaced with P _i

Step 6) is for particle i, if its best learning object inefficacy algebraically t _iSurpass preset threshold value (value was 5 generations among the present invention program), then adopt " Orthogonal Composite mode " to produce a new best learning object, and with t _iBe made as 0;

Step 7) is carried out above step 2 repeatedly), 3), 4), 5) and 6) up to satisfying end condition, then the value of each dimension is corresponding circuit component value in the solution of global optimum position G representative.

In said process, step 6) relates to the method that a kind of employing " Orthogonal Composite mode " produces a new best learning object, it is characterized in that using the method for orthogonal experiment to excavate the individual historical optimum P of particle _iWith the information of the G of population global optimum and be combined into a learning object that can guide particle to evolve to better direction, concrete operation steps comprises:

(1) for particle i, suppose that the circuit component quantity that needs to optimize is D, the dimension that is to say problem to be optimized is D, generates 2 horizontal quadrature tables with D factor, this orthogonal arrage is total

OK.

(2) produce M solution according to orthogonal arrage, separate X for each _j(each dimension x of 1≤j≤M) _Jd(1≤d≤D), if the corresponding value of the capable d row of j is 1 in the orthogonal arrage, then establish x _Jd=p _Id(be that information comes from the individual historical optimum P of particle _i), otherwise the corresponding value of the capable d row of j is 2 in the orthogonal arrage, then establishes x _Jd=g _d(being that information comes from the G of population global optimum).

(3) to all X _j(1≤j≤M) carry out adaptive value to assess, and find out optimum solution, be made as X _b

(4) according to all M solution, calculate the optimal level (just finding out the optimal level of each factor by the factor-analysis approach of orthogonal experiment) of each dimension.

(5) use new solution X of optimal level combination of each dimension that step (4) calculates _p(being called prediction separates), and assessment X _pAdaptive value.

(6) compare X _bAnd X _pAdaptive value, and with O _iBeing set to wherein adaptive value separates preferably.

Description of drawings:

Fig. 1 power electronic circuit synoptic diagram

Fig. 2 is based on the power circuit element optimal design process flow diagram of quadrature study population

Embodiment:

Further the method for invention is described below in conjunction with accompanying drawing.

In Fig. 1, provided the synoptic diagram of the general structure of power electronic circuit.In circuit, there are circuit components such as various resistors, capacitor and inductor.When using quadrature study particle cluster algorithm to optimize these circuit components, the variable that these circuit components need need be optimized as algorithm also is encoded in the particle.The circuit component number of supposing a circuit is D, and then the particle code length of algorithm is D, the corresponding circuit component of each dimension, and have certain span.The span of different circuit components generally by the deviser rule of thumb or the obtained value of element set.

According to the algorithm flow chart of Fig. 2, initialized the time, population is encoded to the individual X=[x that length is D with circuit component ₁, x ₂..., x _D], the parameter that particle cluster algorithm is set is as follows: the inertia weights omega is made as 0.5, and study factor c is made as 2.0, and population scale N is made as 40, and maximum evolutionary generation G is made as 500; Generate N=40 particle at random and form population, the position of each particle i and velocity encoded cine are expressed as X respectively _i=[x _I1, x _I2..., x _ID] and V _i=[v _I1, v _I2..., v _ID]; Wherein D is code length, and is identical with the number of elements of circuit; Position X _iIn each dimension value representation should the dimension circuit component value; The locative situation of change of speed; Assess the adaptive value of all particles, make the historical optimal location P of particle i _i=[p _I1, p _I2..., p _ID] be current location X _i, the position G=[g of global optimum of whole population ₁, g ₂..., g _D] be all historical optimal location P _iIn best that, the best learning object O of each particle i is set simultaneously _i=[o _I1, o _I2..., o _ID] be current individual historical optimal location P _iAnd best learning object inefficacy algebraically t _i=0.

In the adaptive value of each individual X of assessment, corresponding adaptive value function definition is:

$\mathrm{\Φ}\left(X\right)=\underset{R_L=R_{L\_\min },{\mathrm{\δR}}_L}{\overset{R_{L\_\max }}{\mathrm{\Σ}}}\underset{v_{\mathrm{in}}=v_{\mathrm{in}\_\min },{\mathrm{\δv}}_{\mathrm{in}}}{\overset{v_{\mathrm{in}\_\max }}{\mathrm{\Σ}}}[F_1(R_L,v_{\mathrm{in}},X)+F_2(R_L,v_{\mathrm{in}},X)+F_3(R_L,v_{\mathrm{in}},X)]+F_4\left(X\right)$

R in the above-mentioned formula _{L_min}And R _{L_max}, v _{In_min}And v _{In_max}Be respectively the load R among Fig. 1 _LWith input voltage v _InMinimum value and maximal value; δ R _LWith δ v _InBe respectively R _LAnd v _InChange step.F in the formula ₁To F ₄It is the evaluation criterion of formulating according to the circuit design slip-stick artist.F wherein ₁Be used for assessing output voltage v _oError rate; F ₂Be used for the switching reaction v of evaluating system _d, comprise maximum upper punch and Xia Chong, and stabilization time; F ₃Be used for assessing output voltage v _oThe voltage stabilizing situation; F ₄Be used for the dynamic behaviour of evaluating system under the large-signal condition.These evaluation criterias generally all are to be formulated according to practical situations by the slip-stick artist.

In the particle cluster algorithm operational process, in each generation, all constantly adjusted the value of the solution of each particle representative by the renewal of speed and position, makes the circuit component value optimization that it reflects.In the speed of carrying out particle i and position renewal, by its best learning object O _iTo speed V _iUpgrade: for V _iEach the dimension v _Id(1≤d≤D), more new formula is accordingly: v _Id=ω * v _Id+ c * r * (o _Id-x _Id); Wherein r is the random number between interval [0,1].After speed was upgraded, the operation that the position is upgraded was as follows: for X _iEach the dimension x _Id(1≤d≤D) accordingly more new formula be: x _Id=x _Id+ v _Id

Finish after the renewal of particle's velocity and position, another important operation of technology of the present invention is to use the position X after the variation strategy upgrades particle i _iMake a variation, strengthen the diversity of algorithm.Concrete method is: generate a random number between [0,1] at random earlier, if the value that this random number less than variation Probability p m (pm=0.01), selects certain one dimension of particle also will tie up is being set in optional scope at random.

Upgrade in speed, position, and after mutation operation finishes, algorithm upgrades position X after the variation to particle i _iCarry out the assessment of adaptive value, if new fitness function value is than its historical optimal location P _iThe fitness function value better, then with P _iBe set to X _i, and with the best learning object inefficacy algebraically t of particle i _iBe made as 0, otherwise with the best learning object inefficacy algebraically t of particle i _iBe made as t _i+ 1; In addition, judge new P _iWhether more excellent than the position G of global optimum of population, if then G is replaced with P _i

For particle i, if its best learning object inefficacy algebraically t _iSurpass preset threshold value (value was 5 generations among the present invention program), then adopt " Orthogonal Composite mode " to produce a new best learning object, and with t _iBe made as 0; Here " Orthogonal Composite mode " produces the technology of a new best learning object, it is characterized in that using the method for orthogonal experiment to excavate the individual historical optimum P of particle _iWith the information of the G of population global optimum and be combined into a learning object that can guide particle to evolve to better direction, in " summary of the invention " of concrete operation steps in instructions detailed description has been arranged.

Carry out the operation in each generation of particle cluster algorithm repeatedly, by the effect of quadrature learning strategy and variation strategy, make algorithm under the prerequisite that keeps the quick convergence of population, to have strengthened the diversity of population, avoid algorithm to fall into the defective of local optimum easily.In the present invention, the end condition of algorithm is the iteration in 500 generations.After algorithm was finished the iteration in 500 generations, the value of each dimension was corresponding circuit component value in the solution of resulting global optimum position G representative.

Claims (3)

1. the quadrature based on the band mutation operator is learnt the method that particle cluster algorithm is optimized the power electronic circuit element, it is characterized in that avoiding algorithm to fall into the defective of local optimum easily by design quadrature learning strategy and mutation operator, this technical method may further comprise the steps:

(1) parameter of initialization quadrature study particle cluster algorithm, and N particle at random, the position of each particle i and velocity encoded cine are expressed as X respectively _i=[x _I1, x _I2..., x _ID] and V _i=[v _I1, v _I2..., v _ID]; Wherein D is code length, and is identical with the number of elements of circuit; Position X _iIn each dimension value representation should the dimension circuit component value; The locative situation of change of speed; Assess the adaptive value of all particles, make the historical optimal location P of particle i _i=[p _I1, p _I2..., p _ID] be current location X _i, the position G=[g of global optimum of whole population ₁, g ₂..., g _D] be all historical optimal location P _iIn best that, the best learning object O of each particle i is set simultaneously _i=[o _I1, o _I2..., o _ID] be current individual historical optimal location P _i, best learning object inefficacy algebraically t _i=0;

(2) to each particle i, by its best learning object O _iTo speed V _iUpgrade; For V _iEach the dimension v _Id, 1≤d≤D wherein, more new formula is accordingly: v _Id=ω * v _Id+ c * r * (o _Id-x _Id); Wherein ω is that 0.5, c is that 2.0, r is the random number between interval [0,1];

(3) to each particle i, its current location is upgraded in use location more new formula, for X _iEach the dimension x _Id, 1≤d≤D wherein, more new formula is accordingly: x _Id=x _Id+ v _Id

(4) use variation tactful in the position X after the particle i renewal _iMake a variation, strengthen the diversity of algorithm;

(5) to the position X after the particle i renewal variation _iCarry out the assessment of adaptive value, if new fitness function value is than its historical optimal location P _iThe fitness function value better, then with P _iBe set to X _i, and with the best learning object inefficacy algebraically t of particle i _iBe made as 0, otherwise with the best learning object inefficacy algebraically t of particle i _iBe made as t _i+ 1; In addition, judge new P _iWhether more excellent than the position G of global optimum of population, if then G is replaced with P _i

(6) for particle i, if its best learning object inefficacy algebraically t _iSurpass preset threshold value, value was 5 generations among the present invention program, then adopted " Orthogonal Composite mode " to produce a new best learning object, and with t _iBe made as 0;

(7) carry out above step (2), (3), (4), (5) and (6) repeatedly up to satisfying end condition, then the value of each dimension is corresponding circuit component value in the solution of global optimum position G representative.

2. based on the mutation operator described in the step (4) of claim 1, it is characterized in that strengthening by the mode that increases random perturbation the diversity of algorithm; Concrete method is: earlier generate a random number between [0,1] at random, if this random number less than variation Probability p m, this pm=0.01 selects certain one dimension of particle and the value that will tie up is optionally being set in the scope at random.

3. based on the method based on a new best learning object of " Orthogonal Composite mode " generation described in the step (6) of claim 1, it is characterized in that using the method for orthogonal experiment to excavate the individual historical optimum P of particle _iWith the information of the G of population global optimum and be combined into a learning object O that can guide particle to evolve to better direction _i, concrete operation steps comprises:

(1) for particle i, suppose that the circuit component quantity that needs to optimize is D, the dimension that is to say problem to be optimized is D, generates 2 horizontal quadrature tables with D factor, this orthogonal arrage is total OK;

(2) produce M solution according to orthogonal arrage, separate X for each _jEach the dimension x _Jd, 1≤j≤M wherein, 1≤d≤D if the corresponding value of the capable d row of j is 1 in the orthogonal arrage, then establishes x _Jd=P _Id, namely information comes from the individual historical optimum P of particle _i, otherwise the corresponding value of the capable d row of j is 2 in the orthogonal arrage, then establishes x _Jd=g _d, namely information comes from the G of population global optimum;

(3) to all X _j, carry out the adaptive value assessment, and find out optimum solution, be made as X _b

(4) according to all M solution, calculate the optimal level of each dimension, just find out the optimal level of each factor by the factor-analysis approach of orthogonal experiment;

(5) use new prediction solution X of optimal level combination of each dimension that step (4) calculates _p, and assessment X _pAdaptive value;

(6) compare X _bAnd X _pAdaptive value, and with O _iBeing set to wherein adaptive value separates preferably.

Images