WO2019109757A1 - Method for using particle swarm algorithm to optimize power electronic circuit - Google Patents

Abstract

Provided is a method for using a particle swarm algorithm to optimize a power electronic circuit; on the basis of the goal of reducing computational load, the method uses decoupling technology to divide an optimization process into two parts, which are, respectively, power transmission optimization and feedback network optimization of a power electronic circuit. A mutant operator is also introduced into a particle swarm algorithm so as to increase the diversity of the swarm and improve the optimization efficiency of the algorithm. Testing is performed using an optimized design of a step-down buck regulator as an example, thus proving the effectiveness of the method.

Description

Method for optimizing power electronic circuit by using particle swarm optimization Technical field

The invention relates to the field of power electronics and intelligent computing technology, and in particular to a method for optimizing a power electronic circuit by using a particle swarm optimization algorithm.

Background technique

A power electronic circuit is a type of circuit that provides power to a load as a primary purpose. By adjusting the supply voltage or current, the power transmission is effectively controlled, so that the user's load obtains the required power. Power electronic circuits have a wide range of applications, such as mobile electronic devices, televisions and computers. With the development of semiconductor technology and electronic packaging technology, the demand for automation design of power electronic circuits is increasing.

The methods of automatic design and optimization of power electronic circuits are mainly divided into two types: deterministic algorithms and random algorithms. Deterministic algorithms, such as the gradient method and the hill climbing method, are prone to fall into local optimum, resulting in suboptimal component parameter values. Also, some deterministic algorithm performance is too dependent on the choice of the initial search point. The optimization of power electronic circuits is a highly nonlinear problem, so deterministic algorithms tend to be poor. In contrast, the random algorithm can search the solution space extensively, has low dependence on the initial search point, and has high robustness. Therefore, it is more suitable for optimizing and designing the power electronic circuit than the deterministic method.

In recent years, an evolutionary algorithm belonging to a random algorithm has attracted the attention of many researchers. The evolutionary algorithm is characterized by the need for information about the objective function without the need for special knowledge related to the specific problem. It is not constrained by the restrictive assumptions of search space. It does not require assumptions such as continuity and conductivity. It can find global optimal solutions with high probability from discrete, multi-extreme, high-dimensional problems with noise. , is a global optimization method with high robustness and wide applicability. Therefore, evolutionary algorithms are well suited for the design and optimization of power electronic circuits.

The particle swarm optimization algorithm is a heuristic random algorithm, which belongs to a branch of the evolutionary algorithm and works by simulating the predation behavior of birds and fish in the natural world. Because of its clear definition and easy implementation, the particle swarm optimization algorithm has been widely used since its introduction. In particular, in the field of analog circuit design, particle swarm optimization has been used in RF circuit design, on-chip spiral inductor design, microwave filter design, and so on. Compared with other evolutionary algorithms, particle swarm optimization has the advantages of fast convergence, high precision and stable quality, so it is very suitable for solving optimization problems such as power electronic circuit design.

Summary of the invention

The object of the present invention is to solve the above-mentioned drawbacks in the prior art and to provide a method for optimizing a power electronic circuit using a particle swarm optimization algorithm.

The object of the present invention can be achieved by adopting the following technical solutions:

The invention discloses a method for optimizing a power electronic circuit by using a particle swarm optimization algorithm, and applies the particle swarm optimization algorithm to the optimization design of the power electronic circuit. In order to reduce the computational load, the invented particle swarm optimization algorithm does not perform the optimization of the whole circuit as in the traditional method. Instead, the power electronic circuit is first divided into two parts of decoupling, namely power transmission and feedback network, and then optimized separately. . The encoding of the particles is encoded by real numbers, and the value of each dimension corresponds to a parameter of the component to be optimized. The specific steps of using the particle swarm optimization algorithm to optimize the power electronic circuit are as follows:

S1: setting an algorithm parameter for optimizing the power transmission part, and randomly initializing the first generation particle group of the power transmission part within a range of upper and lower limits of the given device value.

S2. Calculate the fitness value of each particle to evaluate the merits of the solution. The fitness function is:

Where, v _in and R _L are the input voltage and load value, respectively, V _{in, max} and V _in,min are the maximum and minimum values of the input voltage, R _L,max and R _L,min are the maximum and minimum values of the load, Δv _in and δR _L are the steps of changing the input voltage and load, respectively, and CP _i represents the i-th particle group individual coding. F _{1 is} used to evaluate the steady state error of the output voltage, F _{2 is} used to evaluate the constraints of the circuit operation, F _{3 is} used to calculate the steady state ripple voltage on the output voltage, and F _{4 is} used to evaluate the inherent properties of the device, such as the overall Price, physical size, etc.

S3. Update the individual historical optimal pBest of each particle according to the fitness value, and the global optimal gBest of all the particles.

S4. Update the velocity and position vector of each particle.

S5. Using the mutation operator to increase the diversity of the group. The specific method is to generate a random number r randomly distributed between 0 and 1 for each dimension of each particle. If r is less than the mutation probability P _m , then the value of the dimension, ie the corresponding device parameter value, is randomly changed within the range of the upper and lower limits of the given device value.

S6. If the end condition of the power transmission part (such as the maximum number of iterations, etc.) is reached, step S7 is performed, otherwise, the process returns to step S2.

S7. Set algorithm parameters for optimizing the feedback network, and randomly initialize the first generation particle group of the feedback network within a range of upper and lower limits of a given device value.

S8. Calculate the fitness value of each particle, and the fitness function is:

Where CF _i represents the encoding of the i-th particle swarm individual. F _{5 is} used to evaluate the steady state error at the output voltage, F _{6 is} used to estimate the maximum overshoot and undershoot, and the settling time of the output voltage during startup, F _{7 is} used to evaluate the stable ripple voltage on the output voltage, F ₈ Used to evaluate the dynamic performance of the circuit when the input voltage and output resistance are disturbed.

S9. Update the individual historical optimal pBest of each particle according to the fitness value, and the global optimal gBest of all the particles.

S10. Update the velocity and position vector of each particle.

S11. In the same way as the power transmission part, the mutation operator is used to increase the diversity of the feedback network.

S12. If the end condition of the feedback network part is reached, the optimization is ended, otherwise the process returns to step S8.

The present invention has the following advantages and effects over the prior art:

The particle swarm algorithm has a simple concept, is easy to implement, and has a fast convergence speed, so it has been widely used. However, in the optimization design of power electronic circuits, due to the complexity of the problem, sometimes the particle swarm algorithm will converge to the local optimal solution, which is caused by insufficient group diversity. Therefore, the present invention introduces a mutation operator in the particle swarm optimization algorithm, which increases the population diversity in the evolution process, and the performance of the particle swarm optimization algorithm for power electronic circuit optimization is improved.

DRAWINGS

1 is a basic structural diagram of a power electronic circuit;

2 is a flow chart of a method for optimizing a power electronic circuit using a particle swarm optimization algorithm disclosed in the present invention;

Fig. 3 is a basic structural view of a step-down regulator.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. It is a partial embodiment of the invention, and not all of the embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts are within the scope of the present invention.

Example

The flow chart of the method for optimizing the power electronic circuit using the particle swarm optimization algorithm disclosed in this embodiment is shown in FIG. 2 .

Particle swarm optimization is an optimization algorithm based on swarm intelligence. Each individual (particle) in the population maintains two vectors during evolution, namely the velocity vector V _ij =[v _i1 ,v _i2 ,K,v _iD ] and the position vector X _ij =[x _i1 ,x _i2 ,K, x _iD ] (the position vector holds the values of the circuit device parameters, corresponding to CP _i and CF _i in the fitness function), where i represents the number of the particle and D is the dimension of the solution, in the power electronic circuit The number of devices to be optimized is represented in the optimized design. The velocity vector of the particle determines its direction and rate of motion, while the position vector embodies the coordinates of the solution represented by the particle in the solution space. The fitness value function is used to evaluate the pros and cons of the particle position, ie the quality of the solution. Particles have the ability of information memory and communication, and each particle maintains its own historical optimal position vector (represented by pBest _i ), that is, in the process of evolution, if the particle reaches a better position of some fitness value , then record the location into the historical optimal vector. In addition, the group also maintains a global optimal position vector (represented by gBest), which is the optimal one of all particles' pBest, which serves to guide the movement of particles to the global optimal region. In each generation, the particle velocity and position update formula is as follows:

V _ij (t+1)=ωV _ij +c ₁ rand ₁ (p _i -X _ij (t))+c ₂ rand ₂ (p _l -X _ij (t)) (1)

X _ij (t+1)=X _ij (t)+V _ij (t+1)

Where ω is the inertia weight, c ₁ and c ₂ are the acceleration coefficients, and rand ₁ and rand ₂ are two random numbers uniformly distributed from 0 to 1.

The basic structure of the power electronic circuit is shown in Figure 1, which includes the power transmission and feedback network. The power transfer section includes I _P resistors, J _P inductors and K _P capacitors; the feedback network section contains I _F resistors, J _F inductors and K _F capacitors. Two vectors are used to represent the passive components in the two parts:

among them,

In the optimization program, Θ _P and Θ _F are optimized separately. The key to optimization is to properly encode the optimization parameters and design the optimization goals into the fitness function of the particle swarm algorithm.

This design encodes Θ _P and Θ _F into real strings CP and CF, respectively, as shown in the following equation. Each bit represents the value of the corresponding component parameter with a real number.

The fitness function functions of the two parts are defined as follows:

Where Φ _P and Φ _F represent the fitness value functions of the power transmission and feedback network portions, respectively. CP _i and CF _i represent the codes of the individual particle groups corresponding to Θ _P and Θ _{F ,} respectively. The objective functions F ₁ and F _{5 are} used to evaluate the steady state error of the output voltage. F _{2 is} used to evaluate the constraints of the circuit operation. F ₃ and F _{7 are} used to calculate the steady state ripple voltage on the output voltage. F _{4 is} used to evaluate the inherent properties of the device, such as overall price, size, lifetime, and the like. F _{6 is} used to estimate the maximum overshoot and undershoot, as well as the settling time of the output voltage during startup. F _{8 is} used to evaluate the dynamic performance of the circuit when the input voltage and output resistance are disturbed.

In the fitness value function Φ _P of the power transmission part, the objective functions F ₁ , F ₂ , F ₃ , F ₄ are respectively designed as follows:

1) F ₁ :

Define a variance accumulation equation E ² to evaluate the approximate degree of the value v' _o and the reference voltage v _ref in the N _s simulation values after the time domain simulation of the output voltage v _o

If the value of E ₂ is small, the steady state error is small and F ₁ is large. The formula F ₁ is defined as follows

Where W ₁ is the maximum value that F ₁ can reach, and W _{2 is} used to adjust the sensitivity of F ₁ to E ² .

2) F ₂ :

Under steady state conditions, some waveforms are subject to constraints. Assuming λ _C,m is the limit of the quantity q _m under the mth constraint, then F _{2 is} defined as

Where N _C is the number of constraints, W _3,m is the maximum value of the mth constraint, and W _4,m determines the sensitivity of the amount considered.

3) F ₃ :

ripple voltage V _o must be on the expected output _{o v,} within close limits ± Δv _{o exp.} The method of measuring the particle group individual CP _i in F ₃ is to calculate the number of simulation points in which the v _o exceeds v _{o, exp} ±Δv _o in the N _S simulation points. F _{3 is} defined as follows

Where W ₅ is the maximum value that F ₃ can reach, W ₆ is the attenuation constant, and Q ₁ is the number of simulation points beyond the allowable sideband. It can be seen that when Q ₁ increases, F ₃ decreases.

4) F ₄ :

This objective function mainly considers some internal factors related to the device, including physical size, device lifetime, and overall price. F ₄ can be expressed as

Among them, Φ _R , Φ _L and Φ _C are functions for measuring different types of devices, respectively. They are defined as follows

Among them, W _7,i , W _8,j and W _9,k are the maximum values that Φ _R , Φ _L and Φ _C can reach respectively. R _i,max , L _j,max and C _k,max are the maximum values of R _i , L _j and C _k , respectively.

In the adaptive value function Φ _F of the feedback network part, the objective functions F ₅ , F ₆ , F ₇ , F ₈ are respectively defined as follows:

1) F ₅ :

Consistent with F ₁ , defined as

2) F ₆ and F ₈ :

During the start or external disturbance, a transient response v _d will appear,

v _d =v _ref -v' _o

F ₆ and F _{8 are} used to evaluate v _d , including 1) maximum overshoot, 2) maximum undershoot, and 3) settling time of the response during startup or disturbance. The basic forms of F ₆ and F ₈ can be expressed as follows

F ₆ =OV(R _L ,v _in ,CF _i )+UV(R _L ,v _in ,CF _i )+ST(R _L ,v _in ,CF _i )

Where N _T is the number of input and load disturbances in the performance test. OV, UV and ST are for the maximum overshoot, undershoot and v _d the maximum setup time minimizing the objective function. They are defined as follows:

Where W ₁₀ is the maximum value that this objective function can reach, M _p0 is the maximum overshoot, M _p is the actual overshoot, and W ₁₁ is the passband constant.

Where W ₁₂ is the maximum value that this objective function can reach, M _v0 is the maximum undershoot, M _v is the actual undershoot, and W ₁₃ is the passband constant.

Where W ₁₄ is the maximum value that this objective function can reach, T _s0 is a constant, T _s is the actual settling time, and W _{15 is} used to adjust the sensitivity. T _{s is} defined as the settling time in which v _d falls into the α±σ% passband, ie

|v _d (t)|≤0.01σ,t≥T _s

3) F ₇ :

F ₇ is the same as the design method of F ₃ in the power transmission section, and the number of simulation points where v _o exceeds v _{o, exp} ±Δv _o is calculated. F _{7 is} defined as follows

The simulation algorithm of the invention is carried out with an optimized design of a buck regulator as an example. The schematic diagram of the buck regulator is shown in Figure 3. The devices to be optimized for the power transmission part are L and C, and the devices to be optimized for the feedback network are R ₁ , R ₂ , R _C3 , R ₄ , C ₂ , C ₃ , and C ₄ , and the parameters of other devices are known. The maximum number of iterations of the particle swarm algorithm is 500. The parameters are as follows:

parameter	Value
Number of particles	30
Inertia weight w	1.2→0.6
Acceleration factor c 1	2.0
Acceleration factor c 2	2.0
Variation probability P m	0.02

In order to compare with the invented algorithm, the genetic algorithm is used to optimize the design of the same circuit. The simulation results of the two algorithms are simulated separately. The results show that the settling time of the simulated output waveform of the particle swarm algorithm is about 5ms, which is shorter than 20ms of the genetic algorithm, and when the voltage or load changes, the circuit optimized by the particle swarm algorithm has less perturbation, that is, the anti-interference ability is more Strong, experimental results prove that the invented particle swarm optimization algorithm is effective in the optimization design of power electronic circuits.

The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and combinations thereof may be made without departing from the spirit and scope of the invention. Simplifications should all be equivalent replacements and are included in the scope of the present invention.

Claims (5)

1. A method for optimizing a power electronic circuit using a particle swarm optimization algorithm, characterized in that the method comprises:

   S1, initializing an algorithm parameter for optimizing the power transmission part, and randomly initializing the first generation particle group of the power transmission part according to the upper and lower limits of the given working parameters of the electronic component;

   S2. Calculate the fitness value of each particle. The fitness function is:

   

   Where CP _i denotes the i-th particle in the particle group, v _in and R _L are the input voltage and load value, respectively, V _{in, max} and V _in,min are the maximum and minimum values of the input voltage, R _L,max and R _L,min is the maximum and minimum of the load, δv _in and δR _L are the steps to change the input voltage and load, F _{1 is} used to evaluate the steady state error of the output voltage, and F _{2 is} used to evaluate the constraints of the circuit operation. , F _{3 is} used to calculate the steady state ripple voltage on the output voltage, and F _{4 is} used to evaluate the inherent properties of the device;

   S3, according to the fitness value, update the individual historical optimal pBest of each particle, and the global optimal gBest of all the particles;

   S4, updating the velocity and position vector of each particle;

   S5. Using a mutation operator to increase the diversity of the group;

   S6, if the end condition of the power transmission part is reached, step S7 is performed, otherwise return to step S2;

   S7. Initializing an algorithm parameter used to optimize the feedback network, and initializing a first generation particle group of the feedback network according to a given device upper and lower limits;

   S8. Calculate the fitness value of each particle, and the fitness function is:

   

   Where F _{5 is} used to evaluate the steady state error at the output voltage, F _{6 is} used to estimate the maximum overshoot and undershoot, and the settling time of the output voltage during startup, F _{7 is} used to evaluate the stable ripple voltage on the output voltage , F _{8 is} used to evaluate the dynamic performance of the circuit when the input voltage and output resistance are disturbed;

   S9. Update the individual optimal pBest of each particle according to the fitness value, and the global optimal gBest of all the particles;

   S10. Update the velocity and position vector of each particle;

   S11. Using a mutation operator to increase the diversity of the feedback network;

   S12. If the end condition of the power transmission part is reached, the optimization process is ended, otherwise the process returns to step S8.
2. The method for optimizing a power electronic circuit using a particle swarm optimization algorithm according to claim 1, wherein said step S11, using a mutation operator to increase the diversity of the particle population, for each dimension of each particle, A random number r randomly distributed between 0 and 1 is generated. If r is smaller than the mutation probability P _m , the value of the dimension is randomly changed, that is, the corresponding device parameter value.
3. The method for optimizing a power electronic circuit by using a particle swarm optimization algorithm according to claim 1, wherein the method for optimizing a power electronic circuit by using a particle swarm optimization algorithm first divides the power electronic circuit into two parts of decoupling, respectively The power transmission and feedback networks are optimized separately.
4. The method for optimizing a power electronic circuit by using a particle swarm optimization algorithm according to claim 1, wherein the encoding of the particles is performed by real number coding, and the value of each dimension corresponds to a parameter of the component to be optimized.
5. The method of optimizing a power electronic circuit using a particle swarm optimization algorithm according to claim 1, wherein the intrinsic properties of the device include an overall price and a physical size.

Images