Optimal Design of Dual Stator--Winding Induction Generator with Variable Speed Based on Improved Genetic Algorithm

Liu Lingshun12, Hu Yuwen1, Huang Wenxin1

1
P P

Nanjing University of Aeronautics & Astronautics , Nanjing 210016, China 2 navy aviation engineering institute, Yantai 264001, China
P P P P

AbstractThe idea of applying the novel dual stator-winding induction generator based on PWM converter to aviation, tank and automobile supply is presented. The minimization to converters reactive power of control winding is the key to optimization of the system. The determination of excited capacitors to minimize the reactive power of control winding under variable speed is given. The calculating capacity of the machine with a diode bridge rectifier load is proposed. The expression between the current of control winding and parameters of generator is derived. To achieve global searching and improve optimization efficiency, the improved real-coded genetic algorithm is presented. The prototype of 18kW, 270V with DC output is designed. The results show this scheme has prominent effect to reduce reactive power of control winding and the capacity of converter can be reduced to 1/3 of output rated power, which makes the dimension of inverter much smaller. This is rather benefiting for the applications to the above isolated supply system. Index Terms Dual statorwinding induction generator, variable Speed, genetic algorithm, optimal design.
. INTRODUCTION

In recent years, the concepts of More Electric Aircraft and All Electric Aircraft, new-style integrating power system (IPS) in marines and All Electric tank have been advanced, which demand the electrical power as the main driving energy in these fields. And the electrical power demand in automobiles has been rising steadily, too. Thus, there is a need for high-power and high-efficiency automotive power generating system to meet the future growth in electrical power demand. Therefore, it is of great necessity to develop 270V (or 42V) high voltage and DC power supply system [1][2][3]. Induction generators are greatly competitive in above isolated applications because of their inherent advantages such as brushless construction, low maintenance and low capital cost [4]. The conventional self-excited induction generator can provide the reactive exciting power through capacitor paralleled with the output terminal. Whereas, an application limitation is the drastic changes of voltage regulation with load and speed variations. Some remedial measures, such as static reactive power compensation or saturation inductor, will inject harmonics and lead large volume, which have limited induction generator with capacitor to apply more widely [5]. And the rapid development of power electronics can improve these limitations, that the converter supplies the required reactive power to the generator, which meets the need of high quality power supply. To effectively minimizing or eliminating the converterinduced harmonics and optimizing system efficiency, dual

stator-winding induction generator scheme with standard squirrel-cage rotor based on excitation converter has been proposed [4][5]. This system is presented schematically in fig.1. Two sets of three-phase windings are embedded in the stator slots, the one referred as power winding with excited capacitors C, supplies power to DC load via a bridge rectifier., the other termed the control winding, or compensating winding, is connected PWM excited converter, which supplies only the reactive power and none of active power. These two sets of windings are wound for the same number of poles, so they share the same working frequency. Both the two sets of windings have none of electrical connection but magnetic coupling exiting, so the harmonic from the converter of control winding has little disturbance on the load, and the electromagnetic compatibility and efficiency have been improved more. Simultaneously, because of some leakage inductance exiting in the control winding, it connects no or low filter inductor. The reliability of the system will be much higher than that of three-phase induction generator system. And this system is especially suitable for operating with rectifier load.
Prime mover Id Cd Converter
C C C

+ Dual stator-winding induction generator

RL

Control winding

Fig. 1. A novel dual stator-winding induction generator system with voltage-source PWM converter

It is well-know that stand-alone supply system such as aviation, automobiles has a wide speed range, how to make this novel system a good use in these fields is imperative. At present the researches on this novel generator system only focus on the control strategy of constant speed [5] [6], however the research on variable speed has not been found. The design of the generator scheme in wide speed range is more complex than that of constant speed, many especial considerations should been taken into. The most important is the demands of volume, weight and cost in aviation or automobiles power supply. The compensated capacity of this system is greatly relative to generator parameters, load and the speed range. Thereby, the operational range of system should be taken into account when we design the inverter and control winding. How to make the capacity of control winding minimum in the
P P

2343

given speed range, accordingly make the converters volume smallest to match the feature of aviation or automobiles, is the key to the optimization of dual stator-winding induction generator system. It just is the problem to be solved in this paper. The optimal design of induction machines represents a rather involved complex, multivariable nonlinear mathematics problem with numerous constraints. Considering that the traditional optimal methods often obtain local optimum and genetic algorithms are credited with the merit of being capable of more often delivering the global optimum [7]. The improved genetic algorithms such as the original parameters of the optimal design being directly put into initial population, adaptive crossover and mutation, excellent individual protect plan are adopted in this scheme. The work is organized as follows. In section , the optimal scheme of excited capacitors and electromagnetic design theories are proposed. In section , optimal mathematical model and improved genetic algorithms are established. In section , an optimal example is given. At last, some conclusions are drawn.
. ELECTROMAGNETIC DESIGN THEORY OF DUAL STATOR-WINDING INDUCTION GENERATOR WITH VARIABLE SPEED

The sign of I s represents the power flowing direction supplied by PWM converter of control winding, which the negative illuminates the inductive reactive power, and the positive represents the capacitive reactive power. Thus the value of capacitor C must make the absolute value of I s minimum in the whole given working region. From calculating and analysis, to satisfy the balance of reactive power at the rating load , the current of control winding is presenting a concave shape curve with descending firstly and ascending afterward in a given speed range (curve 1or 2 of fig.3). The maximal current will occur at the rated speed or the given highest speed is greatly relative to generator parameters and speed ratio. When the leakage reactance is low and speed ratio is small, the maximal current will occur at the rated speed on point A of curve 1; otherwise, it will occur at the highest speed on point D of curve 2. However, inductive I s is always increasing with the speed increasing under light or no-load, which the I s corresponding to the highest speed is the maximal (point E of curve 3).
A D 1 C 2 Rating load curve B

A.

Optimization scheme of excited capacitors

For the variable speed system, the excited capacitors C should be a constant value, and the reactive power demand caused by working conditions changing such as variable speed or load can be compensated by the control winding. The value of excited capacitors will be crucial to minimize the control winding current in a certain wide speed range. On the conditions of the constant capacitors C, and the given machines parameters, speed, load and output power, the control winding current can be uniquely calculated out based on the equivalent circuit. Therefore the control winding can be regarded as a constant current source branch, as fig.2.
LX

Is (A)

f Nf No- load curve 3 E

fr (Hz)

Fig.3. The laws of control winding current ( f is rated frequency of rotor, N is speed ratio )

The capacitor C should be optimized to make the maximum inductive I s of no-load equal to the maximum capacitive I s of rating load as near as possibly in the given speed range. It can be discussed as follows When the rated I s is shaped as curve 1,

RX

Is

Ip

IL

IC C

LL

RL

Im I r Lm Lr Rr s Em Rm

I sE = I sA

(1) (2)

When the rated I s is shaped as curve 2,

Up

ZL

Fig.2. Equivalent circuit of dual statorwinding induction generator

Where, I sA , I sD , I sE is control winding current of point A, D, E, respectively. Accordingly, the control winding current I s will be minimum in the whole speed range and variable load from zero to rated value. This is the optimal scheme of I s via the capacitor C. Given the above, we can draw the conclusion that the independent variables should include the excited capacitor C and the parameters of machine to be optimized simultaneously. B. Theoretical analysis of the control winding current

Where, L x and R x are leakage inductance and resistance of control winding, respectively. L L And R L are leakage inductance and resistance of power winding, respectively. Lr And Rr are leakage inductance and resistance of rotor winding, respectively. Z L is the load impedance.

2344

According to[6], Vector diagram of this generator according to fig.2 is presented as fig 4.

C. calculating capacity of the generator with a diode bridge rectifier load For AC generator with a diode bridge rectifier load, its ratings of voltage and current should refer to those of the DC output. According to these factors such as the adopted phase numbers of rectifier circuit and connecting mode, the wave of voltage and current, the voltage drop of component and circuit, the phase ratings of AC voltage and current can be found. AC capacity P~ and calculating capacity of the generator P ' are respectively expressed as 6 P~ = 3U I
P ' = k E k N 3U I

- Em

UP

IP

Ir

IL

IC

Is

Im

- UP

I r Rr s

I P RL - jI r X r jIP XL

Em

7
' , phase voltage; = k U d

Where, U =

U p + 2 U D k k cx k B

Fig.4. vector diagram of dual stator-winding induction generator

& can be expressed as The excited current I m

U p , DC output voltage; U D Voltage drop of diode;

Im = Ir+ Is I p
accordingly

1 + cos ; 2

I m = I s I r sin I L sin + I c cos

k cx = 2 cos(

2m

) kE =

Ei U

k B = 2m kN

); 2m = 1.02 1.03 , Coefficient of loss.

sin(

In practice, DC output U p depends on line-line

)2 + (I

p XL

)2

Ir =
(

Rr 2 ) + X r2 s

Em I c = U P C . Xm Then the current of control winding can be deduced as (Up + I pRL )2 + (I p XL )2 (Up + I pRL )2 + (I p XL )2 Xr Is = + R Xm ( r )2 + Xr2 s 2 2C XLI p Up + (Up + I pRL )2 + (I p XL )2 (Up + I pRL )2 + (I p XL )2 Im =

voltage, which goes without the third harmonics. Thus, the mean value of commutation coefficient is adopted P ' = k E k N 1.2 Pd 8 Substituting (8) into the main dimension design formula of machine inner diameter Di and axis length
Lm can be found.

D. winding configuration scheme The two sets of windings have many configuration schemes. Owing to the leakage inductance exited in stator winding, some disturbance will appears to the control of the system. Therefore, the choice principle of electrical angle between the corresponding two sets of windings (ax or by or cz) should make the leakage reactance between the d and q axes of different three-phase windings decouple as well. Mutual leakage inductance matrix between the different three-phase windings is expressed as [9]
M M = M M
mlax mlaz mlay

Considering that output voltage U p

5 of power

winding and capacitor C will be constant in the given working region, to reduce I s , the parameters of generator must be optimal designed. Concretely to say, leakage reactance X L , resistance RL of power winding and leakage reactance X r of rotor should be reduced, and excited reactance X m be increased[6]. Where, the influence of X m is more prominent in low speed region, that of leakage reactance is greater in high speed region. And the control winding is designed for higher voltage than the power winding to reduce the inverter current rating which depends on the turns ratio of the two winding sets.

l 12

M M M

mlay mlax mlaz

M M M

mlaz mlay mlax

(9)

Equation (9) transformed into the d-q coordinate become Mlm = Mmlaxcos + Mmlay cos( + 2 3) + Mmlaz cos( 2 3)

Mldq = Mmlaxsin + Mmlaysin( + 2 3) + Mmlazsin( 2 3)

2345

Where, is electrical angle between the corresponding two three-phase windings, generally = 6 . E. working region of magnetic system Due to the wide speed range of aviation, tank and automobile, the design of magnetic system should not only take into account the working condition of rating speed but also that of low speed. Accordingly, it is essential to check up the rated no-load voltage should be gained and deep magnetic saturation not occurs in the lowest speed. Therefore, the working point should setup less than the knee value of magnetization curve .Otherwise the required voltage will not be obtained because of the deep saturation in the lower speed than rating speed.
. THE OPTIMAL MATHEMATICS MODEL AND IMPROVED GENETIC ALGORITHM

as
X = [ Di , h2 , btz , h2 r , btzr , Lm , B g , K i , C ]T

(14)

The real-coded numbers arrange is adopted to reduce the length of chromosome, which is beneficial to improve the convergence and global searching. C. the constraints and penalty algorithm The constraints include performance constraints and variable constraints. In order to prevent the material from saturation, the stator and rotor back irons should be smaller than the saturated flux density. And the heat load, overload capability and the control winding current in the given speed range must be considered as the constraints. Here the constraints of a generator are

A. objective function and fitness function The capacity of control winding has rather strong influence on the capacity and dimension of converter, which is the most key to cost, volume, and weight of the system. Therefore, the capacity of control winding must be taken measures to reduce. According to the optimal theory of section 2.A, the objective function should be the control winding current I s minimum in the given speed range and rated load on the conditions of meeting the constraint of (1) or (2) basically. It is presented as When the rated I s is shaped as curve 1
f ( X ) = min I sA

g 1 ( x ) = B t 1 1 .7 < 0 g 2 ( x) = Be1 1.6 < 0 g 3( x) = B 1.7 < 0 t2 g 4 ( x ) = B e 2 1 .1 < 0 g ( x) = J 20.0 < 0 1 5 g 6 ( x) = AJ 450 < 0 g 7 ( x ) = K t 2 .0 > 0 g 8 ( x) = I sA (orI sD ) I sE < 1

15

(11) (12)

Penalty techniques are used to transform the constrained design problem into an unconstrained problem such that the genetic algorithm can be used to find the optimum solution. Therefore, an exterior penalty function is adopted in the present work to append the constraints to the objective function. In general, the penalty function is defined as follows
min F ( x) = f ( x) +
0 Gi ( x) = gi ( x) gi gi

When the rated I s is shaped as curve 2

G ( x)
i i =1

16
feasible feasible domain domain

Fitness function is the foundation of choosing arithmetic operators in genetic algorithm, which is usually maximization in the genetic operation. Accordingly, the minimizing objective function should be transformed. The fitness function is defined as follows f ' ( X ) = [ f max f ( X )] (13) Where f max is the maximization of f ( X ) in the current evolution generation. Thus the fitness value will not less zero. B. design variables and coded mode Those have heavy influence on the objective function and performances of machine are usually served as design variables. In order to reduce the computer time and simplify the optimal problem, some variables such as the number of slot and poles are fixed during the design procedure, while the inner stator diameter( Di ) ,the stator slot depth ( h2 ) ,the stator tooth width ( btz ),the rotor slot depth ( h2 r ) ,the rotor tooth width ( btzr ) ,the iron core length ( Lm ) ,the gap magnetic density ( B g ),the ratio between stator inner and outer diameter ( K i ), the excited capacitance ( C ) are chosen as the design variables. The independent variables vector is expressed

x x

is isn' t

17 Where f ( x) is objective function. Gi ( x) is the penalty function of the ith constraints. = 1 T

the successive generation such that the penalty term is of the same order of magnitude as the objective function .Here assimilates the idea of simulated annealing, which makes T reduce gradually and increase accordingly. Thus the penalty press to non-feasible solutions will enhance with the evolution proceeding, and the feasible solutions will reach finally. D. the improving of initial population The initial population should be within the range [Xmin, Xmax] of given variables. N real-coded initial solution vectors produced randomly consist of the initial population. In order to enhance optimal efficiency and convergence rate, the original parameters of the optimized generator are directly put into initial population.

2346

E. crossover operator

current generation, generally 0.1, Pm 2 is the smallest The arithmetic crossover strategy is introduced. A mutation probability of the current generation, generally group of numerical values are produced with random in 0.001, f is the fitness value of selected individual. [0, 1], then the superior pair of parents are selected with As a result, those individuals whose fitness value is the roulette wheel selection [10], the selected probability less than mean fitness will have larger crossover and of every individual is expressed as mutation probability; while those have larger fitness value fi 18 than mean value will have lower crossover and mutation pi = M probability. In order to reduce the probability of local fi optimization in the early evolution, the crossover and i =1 mutation probability of the individual with the largest Where M is population size, f i is the fitness value fitness will not be zero, which enlarges the crossover and of the ith individual, pi is the selected probability of the mutation probability of those individuals with excellent performance and prevent them from logjam. ith individual. If the random value is less than the crossover GExcellent individual protection strategy probability, the arithmetic crossover operator is adopted as Excellent individual protection strategy is an effective method to prevent the excellent individual from varying X ' = kX 1 + (1 k ) X 2 19 and losing in the course of genetic operating. Firstly, X '' = (1 k ) X + kX 1 2 compared the fitness between the most elitist individual Where X ' and X '' are new population individuals, of the previous generation with that of the current X 1 and X 2 are the selected old individuals. k is the generation, if the most elitist of the previous generation is random value01. Thus the two new individuals are better than that of the current, the worst one of the current produced. will be replaced by the most elitist one of the previous Crossover probability and mutation probability are generation. Therefore, the most elitist individual is the key to the genetic behavior and performance, which protested from losing and the phenomenon of the most directly influence the convergence rate. The excessive elitist one of new generation being worse than that of large crossover probability will make the larger previous generation will not appear. destroyed probability of the genetic operating, which the individuals with larger fitness value will be destroyed SAMPLE DESIGN quickly. Otherwise, the excessive small crossover The 18Kw, 4-pole, DC voltage of 270V, dual probability will make the searching course slower, even stator-winding induction generator with a diode bridge to logjam. Therefore the adaptive crossover probability rectifier load is optimized as a sample design. The rated with the varying of fitness value is speed is 4000 rpm and the given ratio of the highest speed ' (P c1 P c2 )( f favg) to rating speed is 3:1. The load range is from no-load to ' f > favg P c1 20 rated-load. The efficiency is 90%. P fmax favg c = P The number of stator slot and rotor slot is 48/38 f ' < favg c1 respectively. The power winding is connected by Y type Where Pc1 is the largest crossover probability of double-layer with 5/6 short-pitch, while the control the current generation, generally is 0.9 Pc 2 is the winding is connected by single-layer. The angle and the smallest crossover probability of the current generation, turns ratio between the two sets of winding is 6 and generally is 0.6 f max is the largest fitness value of the 2:1 respectively. The control winding is embedded in the current generation, f avg is the mean fitness value of bottom layer of stator slots. Population size of 28 is chosen and the maximum of each generation, f ' is the larger fitness value of the generations is 40. Optimal results are shown as follows selected two individuals.
F. mutation operator The gene of mutation bit is randomly selected in the range of [0, 9] for the real-coded chromosome. The mutation operator acts as the local search ability with random, and keeps the diversity of generator. This can prevent the early maturation from appearing. The participant mutated individual is determined by the mutation probability, which is adaptive by [10] (P m1 P m2 )( fmax f ) f > favg P m1 fmax favg P 21 m = P < f f avg m1
TABLE I OPTIMAL PARAMETERS

parameters inner stator diameter(mm) stator slot depth (mm) stator tooth width(mm) rotor slot depth(mm) rotor tooth width(mm) iron core length (mm) gap magnetic density (T) ratio between stator inner and outer diameter The excited capacitance (uF )

Initial value 129.3 15.2 3.7 10.9 5.1 197.0 0.6 0.645 80

Optimal value 121.2 14.6 4.1 10.3 4.8 188.6 0.58 0.65 72

Where Pm1 is the largest mutation probability of the

TABLE 2
OPTIMAL PERFORMANCE

Optimal performance Current of control winding(A)

Initial value 27.2

Optimal value 18.1 33.4 92.2 9.4 236 1.52 1.26 1.43 1.01

Ratio between the capacities of control winding to PN (%) Efficiency (%) J 1 A/mm2
P P

46.1 90.9 9.6 229 1.48 1.2 1.38 0.97

AJ A /mm3 Bt1 T
P P P P

Be1 T Bt 2 T Be 2 T

which is much benefit to reduce the capacity, dimension, cost and control of the system. Therefore, this novel dual stator-winding induction generator is suitable for the wide speed range applications such as aviation, automobile. The improved real-coded genetic algorithm is presented. The original parameters of the optimal generator are directly put into initial population, adaptive crossover and mutation is introduced, which have proved the ability of global searching and improving optimization efficiency. The superiority of this novel generator system to traditional three-phase induction generator is the reactive capacity of control winding and its inverter is much smaller than the rated output power in the wide speed range, which provides a new approach to research the isolated supply with variable speed.
REFERENCE [1] A.Emadi and M.Ehsani. Aircraft power systems: Technology, state of the art and future trends [J].IEEE AES Systems Magazine, January 2000, pp.28-32 [2] LikE, Elbaluk,M.David Kankam. Potential Starter/Generator Technologies for Future Aerospace Application , IEEE AES System Magazine, Vol. 12, No.5, 1997,pp.2431. [3] Malakondaiah Naidu, James Walters.A 4-kW 42-V Induction-Machine-Based Automotive Power Generation System with a Diode Bridge Rectifier and a PWM Inverter. IEEE Trans. on Industry Applications, Vol. 39, No.5, September 2003. [4] C.Bansal,T.S.Bhatti,D.P.Kothari. Bibliography on the Application of Induction Generator in Unconventional Energy Systems. IEEE Trans on Energy Conversions, 2003,183,pp.433438. [5] .Lorunfemi Ojo, Innocent Ewean Davidson. PWM-VSI Inverter-Assisted Stand-Alone Dual Stator-Winding Induction Generator. IEEE Trans. on IA, vol. 36, No.6.2000,pp.1604 --1611. [6] Wangdong, Ma Weiming. Research on dual stator-winding induction generator with static excitation regulator.[J].Proceedings of the CSEE., 2003, 23 (7), pp.145--150. [7]. Goldberg D.E, Genetic Algorithms in Search, Optimization and Machine Learning[M]. AddisonWesley Publishing Company, 1989. [8] B A Balagypob,translated by zhu Yaozhong[M].Design of special. AC machines.Beijing:Guofang industry press,1987 [9] R.F.Schiferl, C.M.Ong. Six Phase synchronous machine with AC and DC stator connections[J]. IEEE Trans. on Power Apparatus and System, Vol.PAS-102, No.8, August 1983, pp.2685-2693 [10] Wang Xiaoping,Cao Liming. Genetic Algorithm---theory,application and soft realization[M].Xian: Xian Jiaotong University Press,2002,73-74(in Chinese)

It is obvious that the main optimal dimensions as core length and stator diameters are less than the initial value and the slots are wider and flatter, which make the leakage reactance smaller. The current of control winding in the given speed ratio 3:1 are shown as fig.5. The current of control winding is reduced from 27.2 A to 18.1 A. the capacity ratio between converter of the control winding and rated output power is reduced to 33.4%,which is much benefit for to reduce the capacity ,dimension and cost of the system.

Fig.5. Optimal performance of dual stator-winding induction generator. . CONCLUSION

The optimal design of the dual stator-winding with variable speed is researched in this paper. The expression between the control winding current and parameters of generator is derived. The determining scheme for calculating capacity of the generator with a diode bridge rectifier load is proposed. The space configuration theory of this two winding sets are proposed, which makes the mutual leakage inductance of d-q equivalent circuit as small as possible. The optimization theory of excited capacitors to minimize reactive power of control winding in the variable speed is advanced. The capacity of inverter of control winding is 33.4% of the rated output power in the speed ratio of 3:1. And efficiency increases to 92.2%,

2348