Research on SVG DC-Side Voltage Control Based-on PSO Algorithm

Transcription

1 I.J. Information Technology and Comuter Science, 06, 0, 9-38 Published Online October 06 in MECS (htt:// DOI: 0.585/ijitcs Research on DC-Side Voltage Control Based-on Algorithm Yingwei Xiao School of Electrical Engineering and Automation, Hefei University of Technology, Hefei, 30009, china Abstract The oeration flow of article swarm otimization () is resented, at the same time the algorithm and algorithm are used to find the otimal value of the standard function, simulation results show that the algorithm has better global search erformance and faster search efficiency. The inertia weight decreasing strategy of algorithm is studied, the simulation results show that the concave function decreasing strategy can accelerate the convergence rate of the algorithm. The stability control of the DC-side voltage is very imortant for the static var generator () comensation, but the disadvantages of the traditional PI control are fixed arameters and oor adatability of dynamic resonse, algorithm is introduced to the otimization of PI arameters, so online -PI control and off-line -PI control are obtained, the voltage loo transfer function is used as the controlled object. The simulation results show that the -PI control can satisfy the time varying system of the controlled object with strong adatability. Index Terms, DC voltage stability control,, off-line -PI, online -PI. I. INTRODUCTION In recent years, the wide use of ower electronic devices makes the semiconductor switching equiments increasingly diverse, these devices consume reactive ower and bring additional burden for the ower grid and adverse effects on the ower quality. Therefore, it is imortant and urgent to control the reactive ower in ower grid in order to ensure the ower transmission quality and ensure the ower system to be effective, stable and normal oeration. is a reactive ower comensation device which comosed of inverter, by controlling the on-off of the inverter switching device, continuous injection or absortion of reactive ower to the ower grid, not only can comensate reactive ower, but also can comensate the harmonic[-3]. At resent, the efficiency of in controlling reactive ower, imroving the ower transmission quality and guarantying the ower system stability and other asects has been recognized. Because of its raid develoment, it becomes one of the most imortant reactive ower comensation devices, and is the focus of the research on the direction of reactive ower comensation and harmonic control. The main research contents of this aer mainly includes: () The oeration flow of algorithm is given, and the algorithm and algorithm are alied to the otimization comarison of standard function extreme value[4-6]. () The inertia weight decreasing strategy of is studied. (3) considering that the stability control of the DC side voltage is very imortant to guarantee the effect of comensation, and considering the disadvantages of the traditional PI control with fixed arameters and the oor adatability, the aer introduces the algorithm with global search erformance into the PI arameters otimization[7-9], online -PI control and off-line -PI control are obtained. (4) The aer takes the voltage loo of as the controlled object, and the simulation verification is done for -PI control. II. PRINCIPE OF According to the different choices of DC energy storage device, the circuit can be divided into the voltage tye bridge circuit and the current tye bridge circuit, in fact, so far the used mostly adots the voltage tye circuit, because the voltage tye circuit has a high running efficiency. Therefore, often refers to the voltage tye circuit as the inverter circuit of dynamic comensation devices. The basic circuit structure is shown in Fig.. Fig.. Basic circuit structure of The rincile of is that the AC side of inverter circuit can outut the voltage u fa, u fb, u of reset fc amlitude and frequency through the on-off control of T~ T 6 six semiconductor switching devices shown in Figure. Under the action of the connecting reactance, comensating current i fa, i fb, i are converted, and offset fc reactive current in ower grid, achieve the urose of reactive ower comensation. Coyright 06 MECS I.J. Information Technology and Comuter Science, 06, 0, 9-38

2 30 Research on DC-Side Voltage Control Based-on Algorithm In ractice, the IGBT full control switch is chosen as the semiconductor switching devices. Therefore, the can be equivalent to the controllable amlitude and hase voltage source connected with the ower grid through the connection reactance. III. AGORITHM RESEARCH In 995, Eberhart and Kennedy used the bird oulation behavior model of biologist Heer to roose a article swarm otimization () algorithm. Each individual in the grou follows a very simle three rules of conduct: () Away from the nearest individual, to avoid a collision. () Flight to the target. (3) Flight to the center of the grou. Different from the model roosed by Heer, Kennedy thought birds initially didn t know the target osition, each individual follows a certain rule to evaluate their current osition, and "local otimum" is used as the best osition that it had flown, "globally otimal" is used as the best location that the entire oulation had flown, birds use the two otimal variables to guide their flight, share information with each other until the final flight target osition. Assuming there is a grou of M articles, each article is in a D dimensional search sace, the seed of i article v v, v,, v, can be exressed as vector i i i id i,, M, osition is exressed as vector i i i id x x, x,, x. Then according to the re-set fitness function, the merits of articles are evaluated for their current osition, the osition and seed of the articles are udated according to the fitness value, udated basis is two extreme values[0,]: individual extremum and global extremum. The individual extremum is the best osition of i article to be found at Best,,,. The global resent, denoted as i i id extremum is the best osition to be found for all the articles in the grou, denoted as gbest,,, g g gd, ( gbest is the otimal value of Best ). For the k iteration, the udate rule of article seed and osition in is shown in the formula (): v v c r x c r x k k k xid xid vid k k k k id id id id gd id () In the formula (), d,,, D, d reresents the current search sace dimension, i,,, M, i reresents the article, k reresents the evolution algebra. x : the d dimensional comonent of the osition vector k id k of article i by k iterations. v : the id d dimensional comonent of the seed vector of article i by k iterations. : the d dimensional comonent of global gd extremum gbest. : the id d dimensional comonent of article i individual extremum Best. c, c : acceleration factor, c is used to adjust the articles seed flying to the individual extremum. c is used to adjust the articles seed flying to the global extremum. : inertia weight, adjust its size and can change the strength of the article swarm search ability[], r, r are the random number between [0,]. A. oeration flow The oeration flow of algorithm can be summarized as:. Encoding for the solution sace of roblem Common encoding way includes binary encoding and real encoding. Real encoding is more commonly used way of algorithm encoding, and it directly searches the solution sace.. Determining the objective function of otimization roblem The objective function is to reflect the characteristics of the roblem to be otimized, that is, the fitness value of the articles. The fitness value of articles in the current osition determines the udate degree of the article's osition and seed. 3. Initialization of the arameters in the algorithm () Determining the article osition and seed range, and initializing oulation size, () Initializing the osition information of each article, (3) Initializing the seed information of each article, (4) Initializing inertia factor, learning factor, iteration number and other arameters. 4. Designing article flying model In the algorithm, the flying seed of articles directly affects the ability of the articles to search for the otimal solution, too fast flying seed makes the articles fly over the global extreme oint, and too slow flying seed may make articles falling into the local extreme oint. The design of flying model directly reflects the ability of the articles to use their own memory and share the social information. 5. Determining the termination condition of algorithm The termination condition of algorithm is usually that the maximum number of iterations is satisfied or the fitness of the articles reaches a certain extreme value, also multile termination conditions can be set according to different roblems. 6. Code imlementation According to the above stes, the algorithm is rogrammed, and need to change one or a few arameters of the algorithm in the rocess of debugging in order to Coyright 06 MECS I.J. Information Technology and Comuter Science, 06, 0, 9-38

3 Research on DC-Side Voltage Control Based-on Algorithm 3 search for the otimal solution. Integrated the above stes, the algorithm flow can be summarized as: Ste : initialization of the algorithm arameters, initializing the oulation size, the number of iterations, the inertia weight, the learning factor, the article osition information and seed information et al. Ste : according to the redetermined fitness function, the current fitness values of all articles are calculated. Ste 3: comare the article current osition and individual extremum Best, if the current osition is better, the current osition is used as the article individual extremum Best. Comare the current article individual extremum Best and grou global extremum gbest, if the current osition of the article is better, the individual extreme Best is used as the global extremum. Ste 4: judging whether the termination of the algorithm is satisfied, if satisfied, the algorithm ends, if not satisfied, enter Ste5. Ste 5: each article's osition and seed are udated according to formula (), continue to enter Ste. B. Analysis of standard function simulation. Simulation model In order to show the sueriority of the algorithm, the algorithm and algorithm are simulated in the same condition, and the solving efficiency of the two algorithms is comared. Seeking the maximum value of Rosenbrock function in a certain range: f x x x x x.048 xi.048i,, 00 In the range of.048 x.048i, i (), the function has two local extreme oints, that is.048, f.048, f and resectively, and the latter is the global maximum.. Simulation arameter setting The samling time of simulation ts 0.00s, the individual fitness is the value of Rosenbrock function, the recirocal of the individual fitness is the objective function. The oulation size of algorithm Size 500, termination of evolutionary algebra G 00, crossover robability Pc 0.90, mutation robability P 0.0 :: Size 0.0/ Size, adative mutation m robability, the smaller the adatation degree, the greater the mutation robability, decimal encoding. algorithm also uses the real encoding, grou size, termination of the evolution algebra are same as algorithm, the article seed range is [-0., 0.]. 3. Simulation results Figure shows the objective function values changes with the number of iterations for the algorithm and the algorithm with linear decreasing strategy of inertia weight, it can be seen that the algorithm has a faster convergence seed. Best J 3. x iter Fig.. Otimization comarison between and According to the above simulation arameters, algorithm and algorithm randomly run five times, the five results of are 388.0, , , , resectively. While the five results of are It can be seen that the algorithm has higher search accuracy and faster convergence seed, and has good stability. In order to make the algorithm obtain the strong global search ability at the early, and the fast local convergence seed at the late, according to the formula (3) the inertia weight is reduced linearly. Among them, Min, Max are the minimum and the maximum inertia weight, iter is the number of the current iteration, itermax is the total number of iterations. Max Min Max iter (3) itermax In order to find a better balance between the global search ability and local search ability, two nonlinear inertia weight decreasing strategies are introduced into the aer, namely the concave function decreasing strategy and the convex function decreasing strategy, resectively, as shown in the formula (4) and formula (5): Max Miniter / itermax / Min Max iter itermax Max / (4) Max Min iter itermax Max (5) In order to fully comare the influence on the global search ability and local search ability of algorithm with three kinds of inertia weight decreasing strategies. Selecting, the grou of article swarm is 30, the number of iterations is 000, the minimum value of the ten dimensional Rosenbrock function of formula (6) is searched, and the objective function is the function value. Coyright 06 MECS I.J. Information Technology and Comuter Science, 06, 0, 9-38

4 3 Research on DC-Side Voltage Control Based-on Algorithm 0 f x, x, x 00 xi xi xi i (6) 00 xj 00 j,,, By using the convex function decreasing inertia weight and linear decreasing inertia weight and concave function decreasing inertia weight algorithm, the minimum value of ten dimensional Rosenbrock function is searched resectively, and 50 times are done resectively. Three kinds of inertia weight decreasing strategies are used to obtain the minimum value of the function, the average value of 50 times is: , , The fitness curve of algorithm is shown in Figure 3 under three strategies. It can be seen that the algorithm by using the concave function decreasing inertia weight has been accelerated, the inertia weight is reduced fast in the initial stage of the algorithm, and the convergence seed of the algorithm is accelerated, the otimization accuracy is the highest. The average value of fitness convex function decreasing linear decreasing concave function decreasing Fig.3. Comarison with three kinds of inertia weight decreasing otimization IV. PID PARAMETER TUNING BASED ON The algorithm is used for PID arameter tuning, according to the algorithm tuning rincile and the different ways of imlementation, the tuning methods can be divided into off-line -PID and on-line - PID[3-8]. Figure 4 is the -PID control system structure diagram. rt et algorithm k k PID i ut fitness function controlled object Fig.4. Control system structure of -PID A. Off-line -PID algorithm yt In off-line -PID, algorithm firstly needs offline learning. The osition and seed of the articles are udated within the scoe of sace according to the size of the fitness function, and constantly search for the otimal PID arameters, after the comletion of the search, the global best osition is substituted into the control system. In this aer, the time integral erformance index of the error absolute value is selected as the fitness function, in order to revent the overshoot and ensure the fast system resonse, the fitness function is shown in the formula (7). (7) 0 3 J e t u t dt t u In the formula (7), et reresents the error, ut reresents the outut of PID controller, t u reresents the rise time,,, 3 reresent the weight of the erformance index of each art. The enalty function is introduced into the fitness function to revent the overshoot. Once the overshoot is generated, the overshoot is used as the most imortant one of the fitness function. At this time, the fitness function is: u if e t J e t u t e t dt t (8) In the formula (8), 4, in order to ensure that the adatation value function is the largest roortion after the overshoot, that is, the current target is to suress the overshoot. After determining the fitness function, according to the oeration flow in MATAB rogramming, the concave function decreasing strategy of the inertia weight is used, and the rocess of the off-line -PID is as follows: Ste : initialization of algorithm arameters. The oulation size, the iteration number, the inertia weight, the learning factor, the initial values of PID arameters, the udate seed, the change range of the osition & seed, and each weight value in formula (7) and (8) are initialized. Ste : takes the each current article PID arameters into the controlled object, gets the erformance arameters of corresonding ste resonse, calculates the fitness value of articles according to formula (7) and (8). Ste 3: comarison between the current article osition and individual extreme Best, if the current osition is better, take the current osition as the article individual extreme Best. Comare the current article individual extremum Best with the grou global extremum gbest, if the current article osition is better, take the individual extreme Best as the global extremum value. Ste 4: judging whether the termination condition of the algorithm is satisfied. If satisfied, the algorithm ends. If not satisfied, enter Ste5. Coyright 06 MECS I.J. Information Technology and Comuter Science, 06, 0, 9-38

5 Research on DC-Side Voltage Control Based-on Algorithm 33 Ste 5: according to the formula (), the osition and seed of each article is udated, continue to enter Ste. In order to verify the effectiveness of the off-line - PID algorithm, the two order transfer function of the formula (9) is used as an examle, and the results are comared with the off-line -PID tuning value and offline -PID tuning value. Controlled object: G s s s (9) Setting simulation arameters, the crossover robability and mutation robability of algorithm are 0.9, resectively. In formula (7) and (8),,, 3, are 0.999, 0.00, 00,.0 resectively. 4 The initial inertia weight of algorithm is 0.95, the termination of inertia weight is 0.4, the concave function decreasing strategy is adoted, the acceleration factor c c. The range of roortional gain arameter of algorithm and algorithm is [0,0], the integral gain arameter and the differential gain arameter are [0,], the number of articles is 30, real encoding, 00 generations of iteration converges quickly fast, the otimal value can be found in 5 generations, but algorithm needs 54 generations around to find the otimal value. From the otimization accuracy, the final fitness value of algorithm is 4.0, the final fitness value of algorithm is According to Figure 5 and Figure 6, algorithm and algorithm can obtain better effect in the off-line tuning of PID arameters, the ste resonse curve is basically no overshoot, and has fast rise time. Comared with algorithm, algorithm has higher searching accuracy and better convergence seed, and the off-line -PID has better tuning effect. B. Online -PID algorithm The so-called online -PID tuning method is to tune PID arameters at each samling eriod. For examle in a samling time, select enough articles, calculate each individual's fitness, according to the degree of fitness the magnitude and direction of the article osition and seed udating value are determined, after a certain number of iterations, the otimal PID arameters are obtained at this samling time. In the next samling time, the otimal PID arameters are searched by using the same method. Firstly, the article fitness function is constructed, and the absolute value of the error and the absolute value of error change rate are used as the fitness function of i article: J i errori i de i (0) Best J In the formula (0), is the error of i article,, are the weight, errori i dei is the error change rate. The introduction of the enalty function is to revent the overshoot: iter Fig.5. Adative value iteration curve of and rin,yout rin t/s Fig.6. The ste resonse of the system under the condition of off-line and off-line Figure 5 is the adative value iteration curve of and, Figure 6 is the ste resonse of controlled object by using two algorithms to find the otimal PID arameters. Can be seen that the algorithm 0 00 if errori i J i J i errori i () Online -PID algorithm can be summarized as the following six stes: Ste : Setting samling time, initializing the size of the oulation, the number of iterations, the inertia weight, the learning factor and the initial value of the PID arameters and the udate seed and the weight of the formula (0). Ste : in the k samling time, take the PID arameters of current each article into the controlled object, calculate the error and error change rate, then calculate the fitness values according to formula (0) and (). Ste 3: comare the article current osition and individual extremum Best, if the current osition is better, the current osition is used as the article individual extremum Best. Comare the current article individual extreme Best and the global extremum gbest, if the current article osition is better, the Coyright 06 MECS I.J. Information Technology and Comuter Science, 06, 0, 9-38

6 34 Research on DC-Side Voltage Control Based-on Algorithm individual extreme Best is used as global extremum. The current best osition is used as the PID arameters in the k samling time. Ste 4: judging whether to meet the algorithm termination conditions in this samling time, if satisfied, jum into Ste6. If not satisfied, enter Ste5. Ste 5: follow the formula () to udate the article osition and seed, and then continue to enter Ste. Ste 6: judge whether the final time of algorithm is satisfied or not, if satisfied, the algorithm ends. If not satisfied, k k, enter the next samling time, enter Ste. Also select formula (9) as the controlled object, the system inut is the ste signal, samling time is ms, PD arameters are otimized on-line by algorithm and algorithm. Proortional gain arameter range is [9.0,.0], differential gain arameter range is [0., 0.3], algorithm crossover robability Pc 0.90, mutation P 0.0 :: Size 0.0/ Size, the initial inertia robability m weight of algorithm is 0.95, the termination of the inertia weight is 0.4, concave function decreasing strategy, acceleration factor c c. The oulation size of two algorithms is 0, the generation of evolutionary algebra is 0, use formula (0) and () as the fitness function, and 0.95, Figure 7 is the ste resonse curve by using on-line -PD and on-line -PD to otimize PD arameters, figure 8 resectively reresents the varying curve of two algorithms under the roortional gain k and the differential gain k d versus time. Can be seen from Figure 8, k and k d have similar time varying trend under two kinds of algorithms. () Comared with the off-line -PID, off-line - PID has faster convergence seed and higher searching accuracy. () The ste resonse of the system is almost the same at on-line -PID and on-line -PID tuning. Because the algorithm has no selection, crossover and mutation oerators in algorithm, it only needs to udate the osition and seed information according to the adative value, so the algorithm is simler and faster. (3) Comare figure 6 and figure 7, the ste resonse of on-line tuning is better than that of off-line tuning. k change kd change Time(s) (a) Proortional gain arameter variation rin Time(s) (b) Differential gain arameter variation Fig.8. Parameters variation waveform of online and online rin,yout Time(s) Fig.7. The system ste resonse under the condition of online and online For the two cases of off-line PID tuning and on-line PID tuning, after a comrehensive comarison, the alication of algorithm and algorithm in PID arameters tuning can be concluded as follows: V. RESEARCH ON DC-SIDE VOTAGE CONTRO In theory, by using full control device, which is used to the reactive ower comensation of the ower grid, is not required to set u the energy storage element in the DC side. But in ractice, considering the converter circuit to absorb the fundamental current in addition, but also including harmonic current, which will cause the round return of art energy between the and the ower suly. Therefore, the existence of the DC energy storage elements can guarantee the stability of the converter circuit[9-3], actually the has both reactive ower comensation and low order harmonic comensation function. Coyright 06 MECS I.J. Information Technology and Comuter Science, 06, 0, 9-38

7 Research on DC-Side Voltage Control Based-on Algorithm 35 A. Energy exchange between DC side and AC side The following is an analysis of the energy exchange between the DC side and the AC side of in the reactive ower and harmonic comensation[4-8]. For system, the instantaneous active ower and instantaneous reactive ower of the ower side are reresented by and s q resectively, s and q are used to reresent the instantaneous active ower and instantaneous reactive ower of outut, the instantaneous active ower and instantaneous reactive ower of the load side are reresented by and q. Considering that the load current contains a artial harmonic, and q can be exressed as: q q q () In the formula (), and q reresent the DC comonent of active ower and reactive ower in the load, and q reresent the AC comonent of active ower and reactive ower in the load. When only reactive ower comensation with : At this time: q 0 q qs q q 0 s (3) (4) There is no active energy exchange between the DC side and the AC side, and the DC side voltage should be ket constant. In theory, the DC side does not need energy storage element at this time. When the harmonic is only comensated: At this time: q q qs q q q s (5) (6) At this oint, is needed to rovide active energy, the average value of instantaneous active ower is zero, but there is the AC comonent in, DC side voltage will fluctuate with. When the reactive ower and harmonic are comensated at the same time: At this time: q q qs q q 0 s (7) (8) At this oint, the voltage of the DC side will fluctuate and is similar to that of the harmonic only when it is comensated. In fact, whether it is the comensation of harmonic or reactive ower, the system itself has a loss. The ower loss of mainly includes the loss of the inverter switching device and the magnetic hysteresis loss of the outut inductor, coer loss, etc. In the case of the switching frequency, the ower loss of increases with the increase of the DC side voltage. The above analysis shows that the DC side voltage will change in the working rocess of, in order to guarantee the stability of the comensation erformance and low ower loss, the DC side voltage must be controlled [9-3]. B. Derivation of voltage loo transfer function For the control of DC side voltage, at resent, it is used to add a voltage outer loo on the basic of the current tracking control inner loo. Figure is the basic circuit structure, with this figure as the object, the transfer function of DC side voltage loo is derived. In figure, usa, u sb and u sc reresent the ower grid three-hase voltage, isa, i sb and i sc reresent the grid side three-hase current, ifa, i and fb i reresent the fc outut three-hase current, u, u and fa fb u fc reresent the three-hase voltage of the PWM inverter outut, u dc reresents the caacitor voltage of DC side. The connecting inductor is and the DC side caacitance is C. Assuming the ower grid is three hase equilibrium, the fundamental electric otential of three-hase ower grid is: usa Em cos( t) usb Em cos( t 0 ) usc Em cos( t 0 ) (9) Considering that the switching frequency is high, it is much higher than the grid frequency, only considering the low frequency comonent of the switching function s k a, b, c, the PWM harmonic comonents are k ignored[0], then: sa 0.5mcos t 0.5 sb 0.5m cost sc 0.5mcos t (0) Coyright 06 MECS I.J. Information Technology and Comuter Science, 06, 0, 9-38

8 36 Research on DC-Side Voltage Control Based-on Algorithm In the formula (0), reresents the initial hase angle of the fundamental wave for the switching function, m reresents the PWM modulation ratio m. Three hases grid side current is: isa Im cost isb Im cost 0 isc Im cos t+0 () The other three hases DC current can be described as a switching function: C. DC side voltage control based on In the aer, for the constant system of formula (5) and the time varying system of formula (6), simulation is done resectively to verify the control effect of the traditional PI control system and time varying system. In ractice, v Tci takes 0.08s, DC side caacitance C takes 0.00 F. constant system time-varying system idc saisa sbisb sci () sc Taking formula (0) and () into formula (), the DC side current of the inverter circuit and the grid side current simlified model are obtained as follow: i 0.75mI cos (3) dc According to the formula (3), the voltage loo structure diagram is shown in Figure 9. Without considering the load current i disturbance, the voltage loo transfer function can be exressed as follow: W ov s m 0.75m cos T Cs Cs In the formula (4), s v v ci (4) is the small inertial delay for the samling of DC side voltage, W s T s is the current inner loo equivalent ci ci function, 0.75m cos is a time varying link that reresents the relationshi between the DC side current and the grid side current, and the controlled object is a time-varying system, in most references, 0.75m cos is, the controlled object is simlified reduced to 0.75m as constant system. W ov s 0.75 T Cs Cs v ci (5) In order to comare the erformance of the traditional PI control system with the time varying system and the constant system, the time varying link K 0.sin t is added in the constant system of formula (5). u dc _ v s W " ov s PI 0.75K v (s) W ci Tci Cs Cs Im 0.75mcos Fig.9. voltage loo structure diagram idc i sc (6) u dc Fig.0. The ste resonse of traditional PI control in constant system and time-varying system Figure 0 is the unit ste resonse of the traditional PI control for the constant system and time varying system, k and k are derived from the classical Z-N method, i k , ki The unit ste resonse index of the constant system of formula (5) in traditional PI control is that overshoot =0.8%, the rise time =0.33s, steady outut y=.005. The unit ste resonse index of the time varying system of formula (6) in traditional PI control is that overshoot =.5%, the rise time =0.038s, steady outut y =0.9507~ Obviously, traditional PI control for the constant system has better dynamic resonse characteristics, and the outut basically has no steady-state error. However, when the system arameters are variable, the overshoot is obviously increased, and there is the outut oscillation and steady-state error. For the time varying system of formula (6), the PI arameters are otimized by using the off-line and the on-line algorithm resectively. The initial inertia weight of algorithm is 0.95, the termination of the inertia weight is 0.4, the concave function decreasing strategy is used, acceleration factor c c, the algorithm oulation size is 0, the evolutionary algebra is the 0 generations, seed range is set to [-,], the range of k is [0,0], the range of k is [0,]. Uses the i formula (0) and () as the fitness function, and 0.95, Figure is the unit ste resonse of time-varying system by using three control methods, and the corresonding resonse index is shown in table. Comared with the traditional PI control, the overshoot of off-line -PI control is reduced, the rise time is faster, but its essence is still the PI control, so system outut has oscillation, although steady-state error roblem has eased, but has not yet been resolved. While the unit ste Coyright 06 MECS I.J. Information Technology and Comuter Science, 06, 0, 9-38

9 Research on DC-Side Voltage Control Based-on Algorithm 37 resonse of on-line -PI control has a faster resonse time, almost no overshoot, no outut oscillation, and steady error is almost zero, it has strong adatability for time varying system. traditional PI off-line -PI on-line -PI Fig.. The ste resonse of three control methods in time-varying system Table. The ste resonse erformance index under three control modes traditional PI off-line -PI on-line -PI k k i /% t /s r ~ ~.0354 variation variation VI. CONCUSIONS With the develoment of ower electronic devices, the load form in the ower grid is becoming more and more diversified, which leads to the severe low ower factor. In the field of reactive ower comensation, with fast resonse seed, smooth adjustment of reactive ower, reliable oerating characteristics and high recision becomes the new research hotsot in the field of reactive ower comensation, reresents the develoment direction of reactive ower comensation device. Because the stability control of the DC side voltage is very imortant for the comensation effect, considering the disadvantages of the traditional PID control with fixed arameters and the oor adatability[3-36], the aer introduces the algorithm with global search erformance into the PI arameters otimization. The aer takes the voltage loo as the controlled object, and the voltage loo transfer function is deduced. Combined with algorithm, on-line -PI control and off-line -PI control are adoted, and decreasing strategy of inertia weight function is used. The simulation results show that -PI control can satisfy the situation of time varying system, and has strong adatability. ACKNOWEDGMENT The authors wish to thank my graduate tutor Hongbin y Wu Professor and Benxian Xiao Professor and graduate student Qinglin Zhang and Peng Zhang. REFERENCES [] Jie Tang, An uo, Ke Zhou, Design and Realization Voltage Control for Static Synchronous Comensator, Transactions of China Electrotechnical Society, vol., no.8,.03-06, Aug.006. [] Qing u, Wei Ni, Xiaohui Wang, iyun Zhuang, Study on for reactive ower comensation and harmonic suression based on active current searation, Control and Decision Conference (CCDC), Chinese, [3] Hanson D.J., Woodhouse M.., Horwill C., Monkhouse D.R., Osborne M.M., STATCOM: a new era of reactive comensation, Power Engineering Journal, vol.6, no.3,.5-60, Mar.00. [4] MingFeng Yeh, MinShyang eu, KaiMin Chen, Design of PID Controller Using Hybrid Particle Swarm Otimization, Fuzzy Theory and it s Alications, International Conference, Taichung, , 0. [5] iqing Xiao, Chengchun Han, Xiaoju Xu, Weiyong Huang, Hybrid Genetic Algorithm and Alication to PID Controllers, Chinese Control and Decision Conference, Xuzhou, , 00. [6] Neath M.J., Swain A.K., Madawala U.K., Thrimawithana D.J, An Otimal PID Controller for a Bidirectional Inductive Power Transfer System Using Multi-objective Genetic Algorithm, IEEE Transactions on Power Electronics, vol.9, no.3,.53-53, Mar.04. [7] Ali Shahcheraghi, Farzin Piltan, Masoud Mokhtar, Omid Avatefiour, Alireza Khalilian, Design a Novel SISO Off-line Tuning of Modified PID Fuzzy Sliding Mode Controller, I.J. Information Technology and Comuter Science, vol.6, no.,.7-83, Feb.04. [8] Zweee Gaing, A article swarm otimization aroach for otimum design of PID controller in AVR system, IEEE Transactions on Energy Conversion, vol.9, no., , Feb.004. [9] Wael M. Elmamlouk, Hossam E. Mostafa, Metwally A. El-Sharkawy, -Based PI Controller for Shunt APF in Distribution Network of Multi Harmonic Sources, I.J. Intelligent Systems and Alications, vol.5, no.8,.54-66, Aug.03. [0] Carlos A. Coello Coello, Gregorio Toscano Pulido, Maximino Salazar echuga, Handling Multile Objectives With Particle Swarm Otimization, IEEE Transactions on Evolutionary Comutation, vol.8, no.3,.56-79, Mar.004. [] Adel T., Abdelkader C., A Particle Swarm Otimization Aroach for Otimum Design of PID Controller for nonlinear systems, Electrical Engineering and Software Alications, International Conference, Hammamet,.-4, 03. [] Aziz M.A.A., Taib M.N., Adnan R., The Effect of Varying Inertia Weight on Particle Swarm Otimization() Algorithm in Otimizing PD Controller of Temerature Control System, Signal Processing and its Alications, IEEE 8th International Colloquium, Melaka, , 0. [3] Jalilvand A., Kimiyaghalam A., Ashouri A., Mahdavi M., Advanced Particle Swarm Otimization-Based PID Controller Parameters Tuning, Multitoic Conference, IEEE International Conference, Karachi, , 008. [4] Khandani K., Jalali A.A., Alioor M., Particle Swarm Coyright 06 MECS I.J. Information Technology and Comuter Science, 06, 0, 9-38

10 38 Research on DC-Side Voltage Control Based-on Algorithm Otimization based design of disturbance rejection PID controller for time delay systems, Intelligent Comuting and Intelligent Systems, IEEE International Conference, Shanghai, , 009. [5] Jianhong Qi, Jinda Cai, Error Modeling and Comensation of 3D Scanning Robot System Based on -RBFNN, International Journal on Smart Sensing and Intelligent Systems, vol.7, no., , Feb.04. [6] RongJong Wai, JengDao ee, Kunun Chuang, Real- Time PID Control Strategy for Maglev Transortation System via Particle Swarm Otimization, IEEE Transactions on Industrial Electronics, vol.58, no., , Feb.0. [7] Zakiah Mohd Yusoff, Zuraida Muhammad, Mohd Hezri Fazalul Rahiman, Mohd Nasir Taib, Real Time Steam Temerature Regulation Using Self-tuning Fuzzy PID Controller on Hydro-diffusion Essential Oil Extraction System, International Journal on Smart Sensing and Intelligent System, vol.6, no.5, , June,03. [8] Hidayat, Sasongko, P.H, Sarjiya, Suharyanto, Performance Analysis of Hybrid PID-ANFIS for Seed Control of Brushless DC Motor Base on Identification Model System, International Journal of Comuter and Information Technology, vol., no.3, , Mar.03. [9] Xiaobin Zhang, Yanru Zhong, Otimal Dynamic Hierarchical Control of the STATCOM DC Side Voltage, Proceedings of the CSEE, vol.9, no.33,.60-67, Nov.009. [0] Qiaoo Xiong, An uo, Xiao Wang, Xiaocong i, Fujun Ma, Huagen Xiao, DC Side Voltage Stability Analysis of Cascaded Static var Generator, Automation of Electric Power Systems, vol.38, no.3,.5-9, Mar.04. [] Hanxiang Cheng, Xianggen Yin, ZhiHua Wang, Oeration analysis of dual three-oint, Proceedings of the th IEEE Mediterranean, Electrotechnical Conference, MEECON, vol.3,.4-44, 004. [] i M., Chiasson J.N., Tolbert.M., Caacitor Voltage Control in a Cascaded Multilevel Inverter as a Static Var Generator, CES/IEEE 5th International Conference on Power Electronics and Motion Control, 006, IPEMC, vol.3,.-5, 006. [3] Gang Yao, uan He, Xin Wang, idan Zhou, HuaJun Yu, The study of hybrid modulation based on cascade and its DC control method, 05 IEEE 8th Canadian Conference on Electrical and Comuter Engineering (CCECE), , 05. [4] Kusic G.., Whyte I.A., Three Phase, Steady-State Static var Generator Filter Design for Power Systems, IEEE Transactions on Power Aaratus and Systems, PAS- 03(4),.8-88, Ar.984. [5] Dixon J., Moran., Rodriguez J., Domke R. Reactive Power Comensation Technologies: State-of-the-Art Review, Proceedings of the IEEE, vol.93, no., 44-64, Dec.005. [6] Xiaobin Zhang, Yanru Zhong, Research on the mathematical model and DC-voltage of static var generator, 4th IEEE Conference on Industrial Electronics and Alications, 009, ICIEA, , 009. [7] Guoeng Zhao, Jinjun iu, Xin Yang, Weiwei Gan, Zhaoan Wang, Analysis and secification of DC side voltage in Static var Generator regarding comensation characteristics of generators, Alied Power Electronics Conference and Exosition, 008, APEC 008, Twenty- Third Annual IEEE,.35-39, 008. [8] Zhang Z., Fahmi N.R., Modeling and analysis of a cascade -level inverters-based with control strategies for electric arc furnace (EAF) alication, IEE Proceedings, Generation, Transmission and Distribution, vol.50, no.,.7-3, Feb.003. [9] Xiaobin Zhang, Yanru Zhong, Dynamic Model and Nonlinear Inverse System-PI Control Strategy of Static Var Generator, Journal of System Simulation, vol., no., , Nov.009. [30] Guoeng Zhao, Minxiao Han, Analysis of significant current tracking error and suression of over-shoot current in grid-connected voltage source converter with abc frame control, Power Electronics, IET, vol.7, no.9, , Se.04. [3] Yoshida H., Fukuyama Y., Takayama S., Nakanishi Y., A Particle Swarm Otimization for Reactive Power and Voltage Control in Electric Power Systems Considering Voltage Security Assessment, Systems Man and Cybernetics, IEEE International Conference, Tokyo, vol.6, , 999. [3] Kiyong Kim, Rao P., Burnworth J.A., Self-Tuning of the PID Controller for a Digital Excitation Control System, IEEE Transactions on Industry Alications, vol.46, no.4, , Ar.00. [33] Keyu i, PID Tuning for Otimal Closed-oo Performance with Secified Gain and Phase Margins, IEEE Transactions on Control Systems Technology, vol.3, no.3, , Mar.03. [34] XingJian Jing, i Cheng, An Otimal PID Control Algorithm for Training Feedforward Neural Networks, IEEE Transactions on Industrial Electronics, vol.60, no.6, , Jun.03. [35] Benxian Xiao, Jun Xiao, Rongbao Chen, PID Controller Parameters Tuning Based-on Satisfaction for Suerheated Steam Temerature of Power Station Boiler, I.J. Information Technology and Comuter Science, vol.6, no.7,. 9-4, Jul.04. [36] Farahani, M., Ganjefar, S., Alizadeh, M., PID controller adjustment using chaotic otimisation algorithm for multiarea load frequency control, IET Control Theory & Alications, vol.6, no.3, , Jul.0. Authors Profiles Yingwei Xiao was born in Hefei, China in 993. She has comleted the B.Sc degree in electrical engineering and automation from Anhui Polytechnic University, Wuhu, China in 05. Currently she is still ursuing her master degree in Hefei University of Technology, Hefei, China. Her research interests include automation of electric ower system, micro grid ower flow calculation. Coyright 06 MECS I.J. Information Technology and Comuter Science, 06, 0, 9-38