CN101252308A - Power electronic circuit optimizing method based on paralleling ant cluster algorithm - Google Patents

Abstract

Disclosed is a power electronic circuit optimization method based on parallel ant colony algorithm, relating to the field of intelligent computation and the field of power electronics. Firstly, the power electronic circuit is decoupled into a power transmission part and a feedback network part, then the two parts are respectively optimized through two optimization processes based on ant colony algorithm. In the first optimization process, the parameters of the elements of the power transmission part are adjustable, while the parameters of the feedback network are fixed. In the second optimization process, the parameters of the elements of the power transmission part are not adjustable, while the parameters of the feedback network are adjustable. The two optimization processes are not isolated, the adjustable parameters in one optimization process will be migrated to the other optimization process to serve as the fixed parameters, and the migration strategy is self-adaptive to control. The optimization method is quite suitable for parallel processes of computers and effectively improves optimization speed.

Description

Method of optimization for power electronic circuit based on paralleling ant cluster algorithm

Technical field:

The present invention relates to artificial intelligence and power electronic two big fields, relate generally to the optimization and the design of power electronic circuit.

Technical background:

In recent two decades, small-signal model is extensively applied in the middle of the design of power electronic circuit.In numerous methods, mean state and derivation usage thereof are the most general one.Because switch converters contains an output capacitance, its cut-off frequency is significantly less than switching frequency, so just can obtain the model of a linear time invariant, the power electronic circuit that becomes when being similar on operating point.Use traditional control theory, feedback network part that just can design circuit.Though this method is very simple and easy, it only is applicable to specific circuit and control method usually, and requires that the running of circuit is had comprehensive understanding.In addition, when circuit is converted into Mathematical Modeling, its state variable averages out, and just can not obtain the details of any relevant response wave shape.Under the situation of large-signal, circuit designers is difficult to the response of accurately predicting circuit.Along with the development of power electronic technology, increasing to the demand that circuit automatic generates, the component parameters of optimizing circuit is to reach the performance of appointment.Before 20 years, the automated design engineering of analog circuit begins to occur greatly.These methods comprise, heuritic approach, the algorithm of knowledge base, simulated annealing and other optimization circuit.Traditional optimization method, as gradient method, hill climbing etc. are utilization to some extent all.But these algorithms are absorbed in the local optimum point easily, cause the component parameters of suboptimum, have limited its utilization in complex space.

Ant group algorithm is a kind of novel simulated evolutionary algorithm, finds the solution a lot of combinatorial optimization problems by the behavior of real ant search food source in simulating nature circle.In the natural world, ant carries out interchange between the individuality by release pheromone, can find a path the shortest between ant nest and food.Because positive feedback, concurrency, strong convergence and robustness that ant group algorithm itself has make it that good performance be arranged in combinatorial optimization problem, as traveling salesman problem, scheduling problem, quadratic assignment problem etc.Compare with other first heuritic approach, ant group algorithm has stronger ability of searching optimum and optimizing ability, and the steady quality of separating also has higher search efficiency.In the Automatic Optimal Design of power electronic circuit, can often relate to a large amount of circuit simulation tests, will be very consuming time so in serial computer, carry out single optimization process.Ant group algorithm has natural concurrency, very be suitable for realizing on parallel computer, so paralleling ant cluster algorithm will improve the speed of finding the solution of circuit optimization problem greatly.

Summary of the invention:

On a computer, use ant group algorithm to find the solution the slow excessively problem of power electronic circuit speed for solving.The present invention applies to paralleling ant cluster algorithm in the optimal design of power electronic circuit, the advantage of this method is: the paralleling ant cluster algorithm of proposition is suitable for realizing on parallel computer, therefore can improve the speed of finding the solution of optimization for power electronic circuit problem greatly.

The concrete grammar of utilization paralleling ant cluster algorithm optimal design power electronic circuit is as follows.

With the power electronic circuit decoupling zero is power delivery and feedback network two parts.Wherein power delivery partly comprises I _PIndividual resistance, J _PIndividual inductance and K _PIndividual electric capacity; Feedback network partly comprises I _FIndividual resistance, J _FIndividual inductance and K _FIndividual electric capacity.Use two passive components in vector representation two parts respectively

${\mathrm{\Θ}}_P=\left[\begin{array}{ccc}{\stackrel{\‾}{R}}_P& {\stackrel{\‾}{L}}_P& {\stackrel{\‾}{C}}_P\end{array}\right],$ ${\mathrm{\Θ}}_F=\left[\begin{array}{ccc}{\stackrel{\‾}{R}}_F& {\stackrel{\‾}{L}}_F& {\stackrel{\‾}{C}}_F\end{array}\right]$

Wherein, ${\stackrel{\&OverBar;}{R}}_P=\left[\begin{array}{cccc}{R}_1& {\mathrm{<figure-callout\; id="2"\; label="R"\; filenames="A20081002567200121.png,A20081002567200131.png"\; state="\{\{state\}\}">R</figure-callout>}}_2& ...& R_{I_P}\end{array}\right],$ ${\stackrel{\&OverBar;}{L}}_P=\left[\begin{array}{cccc}{L}_1& {\mathrm{<figure-callout\; id="2"\; label="L"\; filenames="A20081002567200121.png,A20081002567200131.png"\; state="\{\{state\}\}">L</figure-callout>}}_2& ...& L_{J_P}\end{array}\right],$ ${\stackrel{\&OverBar;}{C}}_P=\left[\begin{array}{cccc}{C}_1& {\mathrm{<figure-callout\; id="2"\; label="C"\; filenames="A20081002567200121.png,A20081002567200131.png"\; state="\{\{state\}\}">C</figure-callout>}}_2& ...& C_{K_P}\end{array}\right],$

${\stackrel{\&OverBar;}{R}}_F=\left[\begin{array}{cccc}{R}_1& {\mathrm{<figure-callout\; id="2"\; label="R"\; filenames="A20081002567200121.png,A20081002567200131.png"\; state="\{\{state\}\}">R</figure-callout>}}_2& ...& R_{I_F}\end{array}\right],$ ${\stackrel{\&OverBar;}{L}}_F=\left[\begin{array}{cccc}{L}_1& {\mathrm{<figure-callout\; id="2"\; label="L"\; filenames="A20081002567200121.png,A20081002567200131.png"\; state="\{\{state\}\}">L</figure-callout>}}_2& ...& L_{J_F}\end{array}\right],$ ${\stackrel{\&OverBar;}{C}}_F=\left[\begin{array}{cccc}{C}_1& {\mathrm{<figure-callout\; id="2"\; label="C"\; filenames="A20081002567200121.png,A20081002567200131.png"\; state="\{\{state\}\}">C</figure-callout>}}_2& ...& C_{K_F}\end{array}\right].$

The optimization process of two ant group algorithms of initialization is respectively in order to optimize Θ _PAnd Θ _FWherein in process one, the component value Θ of feedback network part _FFix the component value Θ of power delivery part _PBe adjustable.And in process two, the element of power delivery part is fixed, and the element of feedback network part is adjustable.

On parallel computer, move the optimization process of two ant group algorithms simultaneously.Two processes of ant group algorithm include following step:

1) turns to a series of nominal value with each the element value in the power electronic circuit is discrete, and these nominal values are mapped to an ant group algorithm optimize in the structural map of power electronic circuit.Wherein, each element all uses a tabulation to represent, each node in the table is represented the nominal value that may obtain of this element.Wherein each node also has been endowed certain pheromones, is used to guide the search of ant.

2) pheromones of each node of initialization is an initial value τ _o, initialization N _PIndividual ant is used to search for best element value combination.

3) allow every ant be followed successively by element adjustable in the optimization process and choose a nominal value, thereby form a paths according to the pheromones size on the node.Ant is selected the node C of element j _jThe probability of [i] is

$P_{\mathrm{ji}}={\mathrm{\&tau;}}_{\mathrm{ji}}/\underset{r=1}{\overset{N+1}{\mathrm{\&Sigma;}}}{\mathrm{\&tau;}}_{\mathrm{jr}}$

τ wherein _JiPheromones on i the node of expression element j.N is the node sum of this element, the nominal value number that just may obtain, and what preserve in N+1 node is value on the successive dynasties optimal path, this makes optimal value to produce bigger influence to whole search.In the process of selecting, the selecteed probability of the node that pheromones is big more is just high more, and the selecteed chance of node that opposite pheromones is more little is more little.

4) after all ants have all constructed the path, circuit simulation test is carried out in the path of each ant, and according to the adaptive value of adaptive value function phi calculating path.Adaptive value function in two optimization processes is identical,

$\mathrm{\&Phi;}\left(\mathrm{ph}\right)=[\underset{R_L=R_{L,\min },\mathrm{\&delta;}{R}_L}{\overset{R_{L,\max }}{\mathrm{\&Sigma;}}}\underset{v_{\mathrm{in}}=V_{\mathrm{in},\min },\mathrm{\&delta;}{v}_{\mathrm{in}}}{\overset{V_{\mathrm{in},\max }}{\mathrm{\&Sigma;}}}O{F}_1(R_L,v_{\mathrm{in}},\mathrm{ph})+O{F}_2(R_L,v_{\mathrm{in}},\mathrm{ph})$

$+O{F}_3(R_L,v_{\mathrm{in}},\mathrm{ph})]+O{F}_4(R_L,v_{\mathrm{in}},\mathrm{ph})$

Wherein, ph represents the path that ant is selected.v _InBe input voltage, it is with step-length δ v _InFrom V _{In, min}Change to V _{In, max}R _LWei load value, it is with step-length δ R _LFrom R _{L, min}Change to R _{L, max}OF ₁, OF ₂, OF ₃, OF ₄Respectively as giving a definition.

OF ₁Be used to be evaluated at the steady-state error of output voltage.Define a variance accumulation equation E ₂, in order to assessment output voltage v _oWith reference voltage v _RefAt N _sThe degree of closeness of individual simulated point

${\mathrm{<figure-callout\; id="2"\; label="E"\; filenames="A20081002567200121.png,A20081002567200131.png"\; state="\{\{state\}\}">E</figure-callout>}}_2=\underset{m=1}{\overset{N_s}{\mathrm{\&Sigma;}}}{[v_o\left(m\right)-v_{\mathrm{ref}}]}^2$

If E ₂Value less, then steady-state error is little, OF ₁Can be bigger.Formula OF ₁Be defined as follows

$O{\mathrm{<figure-callout\; id="1"\; label="F"\; filenames="A20081002567200121.png,A20081002567200131.png"\; state="\{\{state\}\}">F</figure-callout>}}_1=K_1{e}^{-{\mathrm{<figure-callout\; id="2"\; label="E"\; filenames="A20081002567200121.png,A20081002567200131.png"\; state="\{\{state\}\}">E</figure-callout>}}_2/{\mathrm{<figure-callout\; id="2"\; label="K"\; filenames="A20081002567200121.png,A20081002567200131.png"\; state="\{\{state\}\}">K</figure-callout>}}_2}$

Wherein, K ₁Be OF ₁The maximum that can reach, K ₂In order to adjust OF ₁To E ₂Susceptibility.

OF ₂Be used to be evaluated at the settling time of output voltage between the starting period, maximum overshoot and Xia Chong.OF ₄Be used for the dynamic property of evaluation circuits when input voltage and output resistance disturbance.During startup or external disturbance, a transient response v will appear _d, wherein

v _d＝v _ref-v _o

OF ₂And OF ₄In order to assessment v _d, comprise 1) and maximum overshoot, 2) maximum dash 3 down) and during startup or disturbance, the settling time of response.OF ₆And OF ₈Citation form can be expressed as follows

OF ₂＝OV(R _L，v _in，ph)+UV(R _L，v _in，ph)+ST(R _L，v _in，ph)

$\mathrm{<figure-callout\; id="4"\; label="O\; F"\; filenames="A20081002567200121.png"\; state="\{\{state\}\}">O</figure-callout>}{F}_{}{}_4=\underset{i=1}{\overset{N_T}{\mathrm{\&Sigma;}}}\mathrm{OV}(R_{L,i},v_{\mathrm{in},i},\mathrm{ph})+\mathrm{UV}(R_{L,i},v_{\mathrm{in},i},\mathrm{ph})+\mathrm{ST}(R_{L,i},v_{\mathrm{in},i},\mathrm{ph})$

N wherein _TBe the input and the number of times of load disturbance in performance test.

In above formula, OV, UV and ST minimize maximum overshoot, maximum dashing down and v _dThe target function of settling time.They are as giving a definition:

$\mathrm{OV}=\frac{K_3}{1+e^{\frac{M_p-M_{p0}}{{\mathrm{<figure-callout\; id="4"\; label="K"\; filenames="A20081002567200121.png"\; state="\{\{state\}\}">K</figure-callout>}}_4}}}$

K wherein ₃Be the maximum that this target function can reach, M _P0Be maximum overshoot, M _pBe actual overshoot, K ₄It is the passband constant.

$\mathrm{UV}=\frac{K_5}{1+e^{\frac{M_v-M_{v0}}{K_6}}}$

K wherein ₅Be the maximum that this target function can reach, M _V0Be to dash M under the maximum _vBe actual following dashing, K ₆It is the passband constant.

$\mathrm{ST}=\frac{K_7}{1+e^{\frac{T_s-T_{s0}}{K_8}}}$

K wherein ₇Be the maximum that this target function can reach, T _S0Be a constant, T _sBe actual settling time, K ₈Be used to adjust susceptibility.T _sBe defined as v _dFall into the settling time of α ± σ % passband.Just,

|v _d(t)|≤0.01σ，t≥T _s

OF ₃With the stable ripple voltage on the assessment output voltage.v _oOn ripple voltage must be at expection output v _{O, exp}Near ± Δ v _oIn the limit.At OF ₃The method of the middle path ch of measurement is to calculate at N _sIn the individual simulated point, v _oExceed v _{O, exp}± Δ v _oThe simulated point number.OF ₃Be defined as follows

${\mathrm{OF}}_3=K_9{e}^{-{\mathrm{<figure-callout\; id="1"\; label="A"\; filenames="A20081002567200121.png,A20081002567200131.png"\; state="\{\{state\}\}">A</figure-callout>}}_1/K_{10}}$

Wherein, K ₉Be OF ₃The maximum that can reach, K ₁₀Be attenuation constant, A ₁Be to exceed the simulated point number that allows sideband.As seen, work as A ₁When increasing, OF ₃Reduce.

5) pheromones on the new node more.Pheromones for the node on the m paths is preferably strengthened, and more new formula is

τ _ji＝α×τ _ji+β×Δτ

Wherein, α is an evaporation coefficient, and β is the plain coefficient that increases of control information, and Δ τ is the unit value that pheromones increases.For best path β=m, second-best path β=m-1, by that analogy.

For all the other relatively poor paths, only the pheromones of node on these paths is evaporated

τ _ji＝α×τ _ji

The pheromones minimum value that is provided with on the node is τ _Min, the pheromones on node is reduced to τ _MinThe time, will no longer this node be carried out the evaporation of pheromones.

6) the successive dynasties optimal path that ant is selected carries out a Local Search renewal at random, and its element value is carried out Local Search in a restricted portion.For i element, its search radius is R _i, when optimal path was carried out Local Search, the value that each element is original was changed into another nominal value in the search radius, and new path is assessed.

If new path is better than original route, then replaces former optimal path, and strengthen the search radius of all nodes with new path ${\stackrel{\&OverBar;}{R}}_i=R_i/\mathrm{shrink},$ Shrink ∈ (0,1) is a zoom factor.

If new path is worse than original route, then keeps original route constant, and reduce the search radius R of all nodes _i=R _i* shrink.

7) continuous repeating step 3) to step 6) to reach the purpose of ant group optimizing.

In the paralleling ant cluster algorithm that proposes, two optimization processes are not what isolate.The adaptive value of the optimal path in k generation is Φ (k) in the suppose process.Owing to taked the strategy of optimum reservation, so Φ (k) 〉=Φ (k-1) is arranged.Calculate each variation, ΔΦ (k)=Φ (k)-Φ (k-1) for Φ (k).If ΔΦ (k)＞0 is then to a migration counter N ⁺Add 1.If N ⁺Greater than a reference value v, illustrate that then this process has been ready to move.When two processes can be moved, with power delivery subelement value Θ adjustable in the process one _PMove in the process two as fixing parameter adjustable feedback network subelement value Θ in the process two _FMove in the process one as fixing parameter.Generally speaking, too frequent migration will make paralleling ant cluster algorithm similar with traditional ant group algorithm effect.And very few migration can cause two processes to be isolated relatively, makes them be absorbed in separately locally optimal solution easily.Therefore, the value of v is the key factor that influences efficiency of algorithm.

V is by comparing the bound Φ of Φ (k) and expectation in the present invention _Max(k) and Φ _Min(k) and self adaptation adjusts.In generation, pass through a some U at k by constructing one ₁: [k-2, Φ _MaxAnd U (k-2)] ₂: [k-1, Φ _Max(k-1)] straight line is predicted the upper limit

The utilization linear extrapolation,

Can be expressed as

${\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right)={\mathrm{\&Phi;}}_{\max }(k-1)+[{\mathrm{\&Phi;}}_{\max }(k-1)-{\mathrm{\&Phi;}}_{\max }(k-2)]$

$=2{\mathrm{\&Phi;}}_{\max }(k-1)-{\mathrm{\&Phi;}}_{\max }(k-2)$

Same, the desired value of lower limit also can be expressed as

${\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right)={\mathrm{\&Phi;}}_{\min }(k-1)+[{\mathrm{\&Phi;}}_{\min }(k-1)-{\mathrm{\&Phi;}}_{\min }(k-2)]$

$=2{\mathrm{\&Phi;}}_{\min }(k-1)-{\mathrm{\&Phi;}}_{\min }(k-2)$

If, ${\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right)>\mathrm{\&Phi;}\left(k\right)>{\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right),$ So

${\mathrm{\&Phi;}}_{\max }\left(k\right)={\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right)$

${\mathrm{\&Phi;}}_{\min }\left(k\right)={\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right)$

Otherwise the utilization linear interpolation is Φ _Max(k) and Φ _Min(k) calculate new value Φ _Min' (k) and Φ _Max' (k).

Φ _max′(k)＝Φ(k)+ΔΦ(k)＝2Φ(k)-Φ(k-1)

Φ _min′(k)＝Φ(k)-ΔΦ(k)＝Φ(k-1)

Based on above method, divide following four kinds of situations to adjust v.

Situation 1: if ${\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right)>{\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right),$ Keep the value of v constant.

${\mathrm{\&Phi;}}_{\max }\left(k\right)=\max [{\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right),{\mathrm{\&Phi;}}_{\max }^{\&prime;}\left(k\right)]$ ${\mathrm{\&Phi;}}_{\min }\left(k\right)=\min [{\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right),{\mathrm{\&Phi;}}_{\min }^{\&prime;}\left(k\right)]$

Situation 2: if ${\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right)>\mathrm{\&Phi;}\left(k\right)>{\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right),$ Keep the value of v constant, ${\mathrm{\&Phi;}}_{\max }\left(k\right)={\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right),$

${\mathrm{\&Phi;}}_{\min }\left(k\right)={\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right).$

Situation 3: if $\mathrm{\&Phi;}\left(k\right)>{\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right)>{\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right),$ The collaborative of this non-adjustable parameter in this optimization process of explanation and adjustable parameter is very high.Adjustable parameter in current process probably can further be optimized.Therefore, the migration between the process should be delayed v=v+1.Simultaneously, Φ _Max(k)=Φ _Max' (k), Φ _Min(k)=Φ _Min' (k).

Situation 4: if ${\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right)>{\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right)>\mathrm{\&Phi;}\left(k\right),$ The collaborative of this non-adjustable parameter in this optimization process of explanation and adjustable parameter is very low.Further optimization possibility is lower for adjustable parameter in the current process.Therefore, the migration v=v-1 between the quickening process.Simultaneously, Φ _Max(k)=Φ _Max' (k), Φ _Min(k)=Φ _Min' (k).

V adjusts in the optimization of whole paralleling ant cluster algorithm adaptively as can be seen.V is controlling the frequency that two processes are moved as adaptive parameter.The migration frequency will not increase when a process has potentiality to do more deep optimization, otherwise the migration frequency can reduce when a process has potential to do more deep optimization.

Description of drawings:

The implementation schematic diagram of Fig. 1 paralleling ant cluster algorithm

Fig. 2 $\mathrm{\&Phi;}\left(k\right)>{\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right)>{\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right)$ The time the computational methods schematic diagram

Fig. 3 ${\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right)>{\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right)>\mathrm{\&Phi;}\left(k\right)$ The time the computational methods schematic diagram

Fig. 4 ${\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right)>\mathrm{\&Phi;}\left(k\right)>{\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right)$ The time the computational methods schematic diagram

Fig. 5 ${\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right)>{\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right)$ The time the computational methods schematic diagram

Embodiment:

The invention will be further described below in conjunction with the drawings and specific embodiments.

The implementation of paralleling ant cluster algorithm as shown in Figure 1.That the band shade is nonadjustable parameter Θ in process one _F, that the band shade is nonadjustable parameter Θ in process two _PThe process of two ant group algorithms is carried out on parallel computer simultaneously, and step is as follows:

1. upgrade the path that ant is selected in the optimization process one according to the step in the ant group algorithm, and calculate adaptive value function phi (k) and ΔΦ (k).If ΔΦ (k)＞0, N ⁺=N ⁺+ 1.Otherwise, N ⁺Remain unchanged.

2. according to formula

${\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right)={2\mathrm{\&Phi;}}_{\max }(k-1)-{\mathrm{\&Phi;}}_{\max }(k-2),$ ${\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right)={2\mathrm{\&Phi;}}_{\min }(k-1)-{\mathrm{\&Phi;}}_{\min }(k-2)$

Calculate respectively

With

3. if $\mathrm{\&Phi;}\left(k\right)>{\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right)>{\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right),$ v＝v+1。According to formula

Φ _max(k)＝Φ _max′(k)＝2Φ(k)-Φ(k-1)，Φ _min(k)＝Φ _min′(k)＝Φ(k-1)

Calculate Φ _Max(k) and Φ _Max(k), as shown in Figure 2.

4. if ${\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right)>{\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right)>\mathrm{\&Phi;}\left(k\right),$ v＝v-1。According to formula

Φ _max(k)＝Φ _max′(k)＝2Φ(k)-Φ(k-1)，Φ _min(k)＝Φ _min′(k)＝Φ(k-1)

Calculate Φ _Max(k) and Φ _Max(k), as shown in Figure 3.

5. if ${\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right)>\mathrm{\&Phi;}\left(k\right)>{\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right),$ Keep v constant, ${\mathrm{\&Phi;}}_{\max }\left(k\right)={\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right),$

${\mathrm{\&Phi;}}_{\min }\left(k\right)={\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right),$ As shown in Figure 4.

6. if ${\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right)>{\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right),$ The same v that keeps is constant.According to formula

${\mathrm{\&Phi;}}_{\max }\left(k\right)=\max [{\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right),{\mathrm{\&Phi;}}_{\max }^{\&prime;}\left(k\right)],$ ${\mathrm{\&Phi;}}_{\min }\left(k\right)=\min [{\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right),{\mathrm{\&Phi;}}_{\min }^{\&prime;}\left(k\right)]$

Calculate Φ _Max(k) and Φ _Max(k), as shown in Figure 5.

7. if N ⁺＜v, process one repeating step 1 to 6.If N ⁺=v, process once in best adjustable parameter be ready to the process of moving to two and suffered.Process is waited for up to process two for a moment and also is ready for migration.Be noted that step 1 to 6 also execution in process two simultaneously.

8. when two processes all are ready to, will carry out the step of migration.The Θ of optimum in process one _PWill move to the fixing parameter of conduct in the process two, and the Θ of optimum in the process two _FThen will move in the process one as fixing parameter.

9. after migration is carried out, N ⁺Can be set to 1, continue as two process repeating steps 1 to 9 then.

At last, if move at certain algebraically N _MaxAfter still do not take place, will carry out an enforceable migration.Can prevent that like this two processes from sinking into their locally optimal solution, optimize processes and become two independently mutual incoherent ant group algorithms.In addition, if N _SMSpontaneous migration does not take place more than inferior, can think that then algorithm has reached optimal solution.In this case, the optimal solution in two processes is exactly to optimize the end value that obtains.

Claims (3)

1, a kind of method of optimization for power electronic circuit based on paralleling ant cluster algorithm is characterized in that, this method is specific as follows: with the power electronic circuit decoupling zero is power delivery and feedback network two parts.The optimization process of two ant group algorithms of initialization.Wherein in process one, the component value of feedback network part is fixed, and the component value of power delivery part is adjustable.And in process two, the element of power delivery part is fixed, and the element of feedback network part is adjustable.Move the optimization process of two ant group algorithms on parallel computer simultaneously, these two optimization processes are not what isolate.The adaptive value of the optimal path in k generation is Φ (k) in the suppose process, calculates each variation for Φ (k), ΔΦ (k)=Φ (k)-Φ (k-1).If ΔΦ (k)＞0 is then to a migration counter N ⁺Add 1.If N ⁺Greater than a reference value v, illustrate that then this process has been ready to move.When two processes can be moved, the optimum component value of power delivery adjustable in the process one part is moved in the process two as fixing parameter, and the optimum component value of feedback network part adjustable in the process two is moved in the process one as the parameter of fixing.

2, based on the described a kind of method of optimization for power electronic circuit based on paralleling ant cluster algorithm of claim 1, it is characterized in that: in paralleling ant cluster algorithm, v is by comparing the bound Φ of Φ (k) and expectation _Max(k) and Φ _Min(k) and self adaptation adjusts.In generation, pass through a some U at k by constructing one ₁: [k-2, Φ _MaxAnd U (k-2)] ₂: [k-1, Φ _Max(k-1)] straight line is predicted the upper limit

The utilization linear extrapolation,

Can be expressed as

${\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right)={\mathrm{\&Phi;}}_{\max }(k-1)+[{\mathrm{\&Phi;}}_{\max }(k-1)-{\mathrm{\&Phi;}}_{\max }(k-2)]$

$=2{\mathrm{\&Phi;}}_{\max }(k-1)-{\mathrm{\&Phi;}}_{\max }(k-2)$

Same, the desired value of lower limit also can be expressed as

${\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right)={\mathrm{\&Phi;}}_{\min }(k-1)+[{\mathrm{\&Phi;}}_{\min }(k-1)-{\mathrm{\&Phi;}}_{\min }(k-2)]$

$=2{\mathrm{\&Phi;}}_{\min }(k-1)-{\mathrm{\&Phi;}}_{\min }(k-2)$

If, ${\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right)>\mathrm{\&Phi;}\left(k\right)>{\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right),$ So

${\mathrm{\&Phi;}}_{\max }\left(k\right)={\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right)$

${\mathrm{\&Phi;}}_{\min }\left(k\right)={\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right)$

Otherwise the utilization linear interpolation is Φ _Max(k) and Φ _Min(k) calculate new value Φ _Min' (k) and Φ _Max' (k).

Φ _max′(k)＝Φ(k)+ΔΦ(k)＝2Φ(k)-Φ(k-1)

Φ _min′(k)＝Φ(k)-ΔΦ(k)＝Φ(k-1)

Based on above method, divide following four kinds of situations to adjust v.

Situation 1: if ${\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right)>{\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right),$ Keep the value of v constant.

${\mathrm{\&Phi;}}_{\max }\left(k\right)=\max [{\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right),{\mathrm{\&Phi;}}_{\max }^{\&prime;}\left(k\right)]$ ${\mathrm{\&Phi;}}_{\min }\left(k\right)=\min [{\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right),{\mathrm{\&Phi;}}_{\min }^{\&prime;}\left(k\right)]$

Situation 2: if ${\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right)>\mathrm{\&Phi;}\left(k\right)>{\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right),$ Keep the value of v constant, ${\mathrm{\&Phi;}}_{\max }\left(k\right)={\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right),$

${\mathrm{\&Phi;}}_{\min }\left(k\right)={\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right).$

Situation 3: if $\mathrm{\&Phi;}\left(k\right)>{\widehat{\mathrm{\&Phi;}}}_{\max }\left(k\right),$ The collaborative of this non-adjustable parameter in this optimization process of explanation and adjustable parameter is very high.Adjustable parameter in current process probably can further be optimized.Therefore, the migration between the process should be delayed v=v+1.Simultaneously,

Φ _max(k)＝Φ _max′(k)，Φ _min(k)＝Φ _min′(k)。

Situation 4: if ${\widehat{\mathrm{\&Phi;}}}_{\min }\left(k\right)>\mathrm{\&Phi;}\left(k\right),$ The collaborative of this non-adjustable parameter in this optimization process of explanation and adjustable parameter is very low.The possibility that adjustable parameter in the current process can further be optimized is lower.Therefore, the migration v=v-1 between the quickening process.Simultaneously, Φ _Max(k)=Φ _Max' (k), Φ _Min(k)=Φ _Min' (k).

3, based on the described a kind of method of optimization for power electronic circuit of claim 1, it is characterized in that: if migration is at certain algebraically N based on paralleling ant cluster algorithm _MaxAfter still do not take place, will carry out an enforceable migration.Can prevent that like this two processes from sinking into their locally optimal solution, optimize processes and become two independently mutual incoherent ant group algorithms.In addition, if N _SMSpontaneous migration does not take place more than inferior, can think that then algorithm has reached optimal solution.In this case, the optimal solution in two processes is exactly to optimize the end value that obtains.

Images