Artificial Intelligence Based Control of a Shunt Active ... - Science Direct

Available online at www.sciencedirect.com

ScienceDirect Procedia Computer Science 92 (2016) 273 – 281

2nd International Conference on Intelligent Computing, Communication & Convergence (ICCC-2016) Srikanta Patnaik, Editor in Chief Conference Organized by Interscience Institute of Management and Technology Bhubaneswar, Odisha, India

Artificial Intelligence Based Control of a Shunt Active Power Filter Jayanti Choudharya, D.K. Singhb, S.N. Vermaa, Kamil Ahmada b

a Department of Electrical Engineering, NIT Patna Department of Electronics &Communication Engineering, NIT Patna

Abstract

With day-by-day increasing significance of Power Electronics, the issue of harmonics being introduced in electrical systems needs to be addressed on a priority basis. Active Filters are very apt solution for reducing harmonics. Here a Shunt Active Power Filter with Artificial Neural Network(ANN) Control and Neuro-Fuzzy Control is proposed. For ANN control the concept of predictive and adaptive control has been used which helps in fast estimation of compensating current. The algorithm for predictive control is new and simulation shows quite good results. The dynamics of the dc-link voltage is utilized in a predictive controller to generate the first estimate followed by convergence of the algorithm by an adaptive ANN (adaline) based network. Again Neuro-Fuzzy control is used which gives still better simulation result © 2016 2016The TheAuthors. Authors.Published Published Elsevier © by by Elsevier B.V.B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Selection and peer-review under responsibility of scientific committee of Interscience Institute of Management and Technology. Peer-review under responsibility of the Organizing Committee of ICCC 2016

Keywords: Adaline, Predictive, Adaptive, THD (Total harmonic distortion), SRF (Synchronous Reference Frame), Fuzzy logic, DC link voltage.

* Jayanti Choudhary. Tel.: +91-9973083169 E-mail address: [email protected]

1877-0509 © 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the Organizing Committee of ICCC 2016 doi:10.1016/j.procs.2016.07.356

274

Jayanti Choudhary et al. / Procedia Computer Science 92 (2016) 273 – 281

1.

Introduction

Although power electronic devices such as thyristors and power transistors have been available for twenty-five years or more, state-of-the-art semiconductor device technology is continually improving and thus providing better, reliable and economical components. These improvements combine with the efficiency and controllability advantages of equipment. New and innovative circuits are being developed at a rapid pace and clearly the trend is toward high penetration level of electrical loads. Unfortunately, there are some problems associated with these new circuits and devices. Unlike conventional loads, they control the flow of power by chopping, flattening, or shaping the otherwise sinusoidal power system voltages and currents .These waveform distortions can cause problems for neighbouring loads, and they tend to have an overall detrimental effect on the quality of electrical “pollution”, whether it is produced by large single-source or by the cumulative effect of many small loads and often propagate along distribution feeders. In fact, distortion is usually amplified at points remote to the sources. sensitive to pollution from other sources, power electronic loads are at the same time villains and victims from a power quality point of view One innovative concept that has great potential is the Active Power Line Conditioner (APLC), also known as Active Power filter. It appears to be an attractive, viable method for reducing voltage and current harmonic distortion, voltage spikes, transients and flicker. It injects equal but opposite distortion thereby cancelling the original problem and improving power quality on the connected power system. The technology of active power filter has been developed for harmonic compensation, reactive power reduction and balancing voltage in an ac power network. Active power filter dynamic performance depends mainly on how fast and accurate harmonics are eliminated from the load current. So many harmonic elimination techniques are available to us. As given in references fast estimation of current compensating is done by using methods like d−q theory presented in [5], p−q topology as in [6], [7], [8], concept of adaptive filters as said in [9], wavelet technology [10], genetic algorithm method (GA) and artificial neural network (ANN) etc. Here, combinations of predictive and adaptive controller techniques which will give faster responses have been adopted. In this there is a need of 2 ANN controllers. Whenever the change in load is detected, first of all predictive controller will do its work in part of compensating current as fast as possible in parallel with change in voltage across capacitor. Next adaline based controller will start working for faster steady state value. Single phase shunt APF is shown in below figure.1

Fig.1: Single phase shunt APF 2. Reference Current Estimation[1] APF compensation is shown in Fig. 1. The expression of the load current is, (1) iL(t) = iα1+iβ1+ih ݅ߙ1 is in-phase component and ݅ߚ1 is quadrature component of phase current at fundamental frequency. Except these, all the remaining included in ݄݅. The in phase component of the load current and the corresponding per-phase source voltage are expressed as (2) ݅α1(t) = Iα1 cos ߱t

275

Jayanti Choudhary et al. / Procedia Computer Science 92 (2016) 273 – 281

vs(t) = ܸm cos ߱t (3) The compensating current becomes ݅c (t) = ݅L (t) −݅α1(t) (4) ݅c (t) = ݅L (t) −Iα1 cos ߱t The compensation of harmonics and reactive power is done by APF. Where, 1is the peak magnitude of the in phase current. The reference current for the APF is easily be set as per (4) once ‫ߙܫ‬1 estimation is over. i(t) is measured using current sensors. ‫ߙܫ‬1 will be find out in two steps. First ‫ߙܫ‬1 value should be find by a single layer ANN based algorithm. Next to get quick convergence adaline based ANN will be used. To maintain dc-link voltage, inverter draws a small current (݅௦ఈ (t)). 3. APF DC Link Voltage Control [1] The circuit that has been used for providing the compensating current is actually a voltage source inverter. The ac side is connected to the three phase power system. On the DC side is a capacitor called DC link capacitor. The dc link capacitor voltage changes whenever there is a change in the load. To get a desired regulated voltage a controller is used. For this the capacitor draws current ݅sα which is in-phase with the source voltage. From the power balance equation, ௗ௏ ܲௗ௖ ൌ ‫ܥ‬ௗ௖ ܸௗ௖ ೏೎ (5) ௗ௧ Where, Pdc is power required to maintain the voltage vdc across the link. By applying small changes in ݅ܿ, ݅‫ݏ‬, ܸ݀, and ‫ ݏݒ‬near the operating point, new set of variables are derived, (6) ݅ܿ (t) = ‫ܫ‬c + Δ݅ܿ (7) ݅sα(t) = ݅sα + Δ݅sα (8) vdc(t) = ܸ݀c+ Δ‫݀ݒ‬c (9) ‫( ݏݒ‬t) = ܸ‫ ݏ‬+ Δ‫ݏݒ‬ Where,c, ‫ܫ‬s, and Vs are the rms values and Vd the dc value of the quantities. After some modifications, ீమ ሺ௦ሻீయ ሺ௦ሻ (10) ο‫ܫ‬௖ ሺ‫ݏ‬ሻ οܸௗ௖ ሺ‫ݏ‬ሻ ൌ ଵା௄ீభ ሺ௦ሻீమ ሺ௦ሻீయ ሺ௦ሻ

“This shows the possibility of estimation of the compensating current from the change in Vd. The PI controller gains are used to regulate the dc link voltage which is governed by the following two inequalities.” ‫ܫ‬ௌ ൏

‫כ‬ ஼೏೎௏೏೎

(11)

ଷ௄೛ ௅೑ ௄ು ௏௦

‫ܫ‬ௌ ൏ ଶ௄

೛ ோ೑ ା௅೑ ௄಺

(12) ‫ܫ‬s the source current, ܸ݀c‫ כ‬is the reference dc-link voltage and ܸ‫ ݏ‬is the source voltage. To design the starting guess of ‫ܲܭ‬and ‫ ܫܭ‬and also to set their limits, equations (11) and (12) are used. 4. Uncompensated System In uncompensated system three-phase supply is directly connected to nonlinear load. No filter is connected here. A scope is for showing source and load currents. Waveforms are source current and load current respectively. Source current is not sinusoidal in this case.

FiFig 2: Simulation results for uncompensated system (time in seconds)

276

Jayanti Choudhary et al. / Procedia Computer Science 92 (2016) 273 – 281

5. Systems with Conventional SRF Figure 3 shows the simulation diagram for system with basic conventional filter. Three single phase voltage sources are connected to the nonlinear load through APF as shown in figure. Source current and voltages are first converted from abc-αβ using Clark transformation and then αβ- dq using park’s transformation. After that again using reverse transformation we get reference current signals which will transform to hysteresis controller to get pulses, which are required for switching of IGBT’s in APF. For managing the shunt APF in real time, instantaneous active power (p-q) theory is used. Those equations are as follows, ‫ܫ‬௔ ‫ܫ‬ ଶ ܿ‫•‘… Ͳʹͳݏ݋ܿ Ͳݏ݋‬ሺെͳʹͲሻ ൨ ൥‫ܫ‬௕ ൩ ൤ ఈ൨ ൌ ට ൤ (13) ‫ܫ‬ఉ ଷ ‫Ͳ݊݅ݏ‬ ‫‹• Ͳʹͳ݊݅ݏ‬ሺെͳʹͲሻ ‫ܫ‬௖ ‫ܫ‬ ‫ܫ‬ ൤ ௗ ൨ ൌ ቂ ܿ‫ ߠ݊݅ݏ ߠݏ݋‬ቃ ൤ ఈ ൨ (14) ‫ܫ‬௤ ‫ܫ‬ െ‫ ߠݏ݋ܿ ߠ݊݅ݏ‬ఉ ‫ܫ‬ ‫ܫ‬ ൤ ௗௗ௖ ൨ ൌ ቂ ܿ‫ ߠ݊݅ݏ ߠݏ݋‬ቃ ൤ ௗ ൨ (15) ‫ܫ‬௤ௗ௖ െ‫ܫ ߠݏ݋ܿ ߠ݊݅ݏ‬௤ ‫ܫ‬௔௥ ܿ‫Ͳݏ݋‬ ‫Ͳ݊݅ݏ‬ ‫ܫ‬ ଶ •‹ ͳʹͲ ൩ ൤ ௗௗ௖ ൨ ൥‫ܫ‬௕௥ ൩ ൌ ට ൥ ܿ‫Ͳʹͳݏ݋‬ (16) ‫ܫ‬ ଷ …‘•ሺെͳʹͲሻ •‹ሺെͳʹͲሻ ௤ௗ௖ ‫ܫ‬௖௥ ࡵࢇ࢘ ࡵࢇ ࡵࢇࢋ ൥ࡵ࢈ࢋ ൩ ൌ ൥ࡵ࢈࢘ ൩ െ ൥ࡵ࢈ ൩ (17) ࡵࢉࢋ ࡵࢉ࢘ ࡵࢉ The simulation results are shown in figure 4. Waveforms are of the input current, output current and compensation current respectively. Source current is not pure sinusoidal in this case. So some advanced techniques are implemented to overcome this problem with APF.Total harmonic distortion (THD) is reduced up to 6.12 % with this filter

Fig 3: Simulation diagram for system with conventional SRF

fig 4: Simulation result for system with conventional SRF (time in seconds)

277

Jayanti Choudhary et al. / Procedia Computer Science 92 (2016) 273 – 281

6. Predictive Controller Based APF Predictive algorithm ensures faster reference generation and is also accountable for better regulation of dc-bus voltage. Ve(t) = 200−Vdc (18) ௄ (19) ‫ )ݐ(כܼ݇( = ݌ܭ‬+ ௓ು) ௄

‫ܭ‬i = (ܼ݇‫’ܸכ‬e(‫ )ݐ‬+ ௓೔ )

(20)

௒

Where, Zk = Y−௓ Y = ‫)ݐ(כ݌ܭ‬+‫ܭ‬i‫’ܸכ‬e(‫)ݐ‬ ூ ‫ܫ‬α = ഀ +ܻ ௓ ‫ܫ‬α = ‫ܫ‬α‫כ‬୚

୚౗ౘౙ౩

(21) (22) (23)

ౣ౗౮ሺ౟౤ሻ

To ensure faster corrective action, the system state is chosen as error voltage and its gradient.[1] u = [ܸ݀‫]݀ݒכ‬ The states ‫ݔ‬1 and ‫ݔ‬2 are designed with the help of the state generator as follows, డ௫ ‫ݔ‬2= భ ‫ݔ‬1 =ve (ߢ) ; డ௞ Where Ve(k)=Vdc*Vdc(k) z(κ), the output error is represented below as, Z(κ) = ‫ݒ‬0(ߢ)- ‫ ߢ(݁ݒ‬− 1) The output ‫ݒ‬0(ߢ) is given to the output state to approximate Iα1 Predictor Training: The network is trained using a large number of test data consisting of inputs and targets.

Fig 5: Simulation diagram for predictive controller based APF

(24) (25)

(26)

278

Jayanti Choudhary et al. / Procedia Computer Science 92 (2016) 273 – 281

Fig 6: Simulation results for predictive controller based APF(time in seconds) The figure 6 shows waveforms for controller based on APF. The source, load and compensation currents are depicted. Source current is not pure sinusoidal. Total harmonic distortion (THD) is reduced up to 5.02 % with this filter. 7. Adaptive Controller based APF Adaline based system is supplied with current samples. This helps in setting a reference to APF. Adaline is designed in order to obtain a minimized THD. Uncompensated source current sample S(k) is given by,[1] ܵሺ݇ሻ ൌ ‫ܫ‬ఈଵ …‘•ɘ–• ൅ ‫ܫ‬ఉଵ •‹ ɘ–• ൅ σோ௡ୀଶ ‫ܫ‬௡ …‘•ሺ݊ɘ݇‫ݐ‬௦ ൅ ‫׎‬௡ ሻ (27) Where, is the step size in discrete domain. The square of error terms for ‫ܭ‬t݄ sample may be expressed as, ൛௦ మ ሺ௞ሻିଶ௦ሺ௞ሻ௔ሺ௞ሻൟ

ߝ ଶ ሺ݇ሻ ൌ ቂ

ଶ

൅ ͳቃ

௔ మ ሺ௞ሻ ଶ ൌ ‫ןܫ‬ଵ ሺ݇ሻܿ‫ ݏ݋‬ଶ ሺɘ݇‫ݐ‬௦

Where ,ƒ ሺ݇ሻ ൅ Equation (20) can also be represented as

(28) ଶ ‫ܫ‬ఉଵ ሺ݇ሻ‫݊݅ݏ‬ଶ ሺɘ݇‫ݐ‬௦ ሻ

ഥ ሺ௞ሻ‫ן‬ ഥ ೅ ሺ௞ሻ௑ሺ௞ሻሿሽ ሼ௦ మ ሺ௞ሻିଶ௦ሺ௞ሻሾ௑ ೅ ሺ௞ሻ‫ן‬

ߝ ଶ ሺ݇ሻ ൌ ቂ

ഥ ሺ௞ሻ‫ן‬ ഥ ೅ ሺ௞ሻ௑ሺ௞ሻ ௑ ೅ ሺ௞ሻ‫ן‬

ቃ

(29)

(30)

ഥ ሺ݇ሻ ൌ  ‫ܫ‬ఈଵ ሺ݇ሻ‫ܫ‬ఉଵ ሺ݇ሻ Where, the vector‫ן‬ i/p vector(k)= ሾ…‘•ɘ–•ǡ •‹ɘ–•ሿT The current required for compensation is then obtained by minimizing the THD which in this case is the error function. This is done using equation (4). This method is not as sluggish as the existing schemes based on adaline as comparatively low number of tuning blocks are required in this case. The input vectors in this case are orthogonal to each other so the calculation part becomes much simple.

Fig 7: Simulation results for adaptive controller based APF(time in seconds)

Jayanti Choudhary et al. / Procedia Computer Science 92 (2016) 273 – 281

279

Total harmonic distortion (THD) is reduced up to 3.02 % with this filter. 8. Predictive and adaptive controller based APF Waveforms are source current, load current and compensation current respectively. Source current is almost sinusoidal. Total harmonic distortion (THD) is reduced up to 1.52 % with this filter.

Fig 8: Circuit Diagram for APF based on predictive and adaptive controller

Fig 9: Output result of the controller APF(time in seconds) 9. Neuro-fuzzy controller based APF Further a neuro-fuzzy system has been developed which upon simulation gives a THD less than 1%.In the neural part adalines are used in the same manner as in the case of adaptive control. Fuzzy Logic The fuzzy logic controller used here has two input membership functions, input1 and input2. input1 is the deviation(Ve ) of dc link voltage from a set value while input2 is the derivative(dVe /dt). Figure 11 shows the input membership functions. The output membership functions taken are out1mf1, out1mf2,….., out1mf25. The base rules discribing the output for the given two inupts are shown in table 1.

280

Jayanti Choudhary et al. / Procedia Computer Science 92 (2016) 273 – 281

The output is multiplied with the three source voltages and a predetermined constant value. This outcome and the output of neural part are averaged and given to a hysteresis controller. Waveforms are source current, load current and compensation current respectively. Source current is almost sinusoidal. Total harmonic distortion (THD) is reduced up to 1.52 % with this filter

Fig. 11 Input Membership Functions Table-1 Base Rule Input 1 1mf1

1mf2

1mf3

1mf4

1mf5

2mf1

out1mf1

out1mf6

out1mf11

out1mf16

out1mf21

2mf2

out1mf2

out1mf7

out1mf12

out1mf17

out1mf22

2mf3

out1mf3

out1mf8

out1mf13

out1mf18

out1mf23

2mf4

out1mf4

out1mf9

out1mf14

out1mf19

out1mf24

2mf5

out1mf5

out1mf10

out1mf15

out1mf20

out1mf25

Input 2

Fig. 11 Input Membership Functions

Fig.12 Output result of the neuro-fuzzy controlled APF (time in seconds)

281

Jayanti Choudhary et al. / Procedia Computer Science 92 (2016) 273 – 281

10. Conclusion For the improvement of converging concept and for the reduction of computational requirement, a combination of adaptive and predictive based ANN controller is depicted for a shunt-type APF. For the regulation of the dc-link voltage in the APF, an algorithm derived from an ANN based PI controller, predictive in nature is used. After that, followed by an adaline-based THD minimization technique is used. The convergence becomes very fast by using only two input vectors followed by two weights. Table-2 THD values of different controllers SRF Predictive Adaptive Predictive and Neuro-Fuzzy Controller Controller Adaptive Controller Controller 6.12%

5.02%

3.02%

1.52%

Less than 1%

11. References [1] A. Bhattacharya and C.Chakraborty, “A shunt active power filter with enhanced performance using ANN-based predictive and adaptive controllers” IEEE Trans. Ind. Electron., vol.58, no.2, pp.421-428, feb.2011. [2] Y. M. Chen and R. M. O’Connell, “Active power line conditioner with a neural network control,” IEEE Trans. Ind. Appl., vol. 33, no. 4, pp. 1131–1136, Jul./Aug. 1997. [3] A. Bhattacharya, C. Chakraborty, and S. Bhattacharya, “Current compensation in shunt type active power filters,” IEEE Ind. Electron. Mag.,vol.3,no. 3, pp. 38–49, Sep. 2009. [5] S. Bhattacharya, T. M. Frank, D. M. Divan, and B. Banerjee, “Active filter system implementation,” IEEE Ind. Appl. Mag., vol. 4, no. 5, pp. 47–63,Sep./Oct. 1998. [6] H. Akagi, Y. Kanazawa, and A. Nabae, “Instantaneous reactive power compensators comprising switching devices without energy storage components,”IEEE Trans. Ind. Appl., vol. IA-20, no. 3, pp. 625–630,May 1984. [7] F. Z. Peng, G.W. Ott, Jr., and D. J. Adams, “Harmonic and reactive power compensation based on the generalized instantaneous reactive power theory for three phase four wire system,” IEEE Trans. Power Electron.,vol. 13, no. 6, pp. 1174–1181, Nov. 1998. [8] R. S. Herrera and P. Salmeron, “Instantaneous reactive power theory: A reference in the nonlinear loads compensation,” IEEE Trans. Ind. Electron.,vol. 56, no. 6, pp. 2015–2022, Jun. 2009. [9] H. Karimi, M. Karimi-Ghartemani, and M. R. Iravani, “An adaptive filter for synchronous extraction of harmonics and distortions,” IEEE Trans.Power Del., vol. 18, no. 4, pp. 1350–1356, Oct. 2003. [10] M. Forghani and S. Afsharnia, “Online wavelet transform-based control strategy for UPQC control system,” IEEE Trans. Power Del., vol. 22, no. 1, pp. 481–491, Jan. 2007. [11] J. H. Marks and T. C. Green, “Predictive transient-following control of shunt and series active power filters,” IEEE Trans. Power Electron.,vol. 17, no. 4, pp. 574–584, Jul. 2002. [12] M. Routimo, M. Salo, and H. Tuusa, “A novel simple prediction based current reference generation method for an active power filter,” in Proc.IEEE APEC, 2004, pp. 3215–3220. [13] B. Widrow and M. A. Lehr, “30 years of adaptive neural networks: Perceptron, madaline and backpropagation,” Proc. IEEE, vol. 78, no. 9, pp. 1415–1442, Sep. 1990.