Utilizing Hydro-Energy to Supply Power in Remote Areas

Abstract

This paper features the performance analysis of an improvised electronic load controller (ELC) for a three-phase Self- Excited Induction Generator (SEIG). A SEIG lags the inherent property of voltage and frequency regulation with varying load demands, thereby controllers are used for the purpose. Levenberg-Marquardt and Scaled Conjugate Gradient are two algorithmic variants of Artificial Neural Network (ANN) which have been used here as load controllers. A comparison of the controlling capability of the two techniques is also featured. Simulation has been carried out in MATLAB and it is found that the results are well acceptable.

Keywords: Self- Excited Induction Generator; short circuit; electronic load controller; Artificial Neural Network.

I. INTRODUCTION

In remote areas harnessing electrical energy from local resources such as wind, bio, hydro and solar energy is very much in use. Micro and Pico hydro systems provide constant power due to constant head and discharge feeding the hydro turbines [1]. Thereby a constant power is generated which needs suitable control techniques.

Get quality help now

writer-Charlotte

Verified writer

Proficient in: Electricity

4.7 (348)

“ Amazing as always, gave her a week to finish a big assignment and came through way ahead of time. ”

+84 relevant experts are online

Hire writer

Self-Excited Induction Generators are found to be attractive for such applications [2-5]. Some of the advantages of SEIGs are - high mechanical strength and ruggedness of the induction machine, reduced short-circuit risks, brushless, low initial and maintenance costs. These advantages of SEIGs have made it superior compared to the conventional synchronous generators.

For maintaining constant voltage and hence constant power at the generator terminals at varying load demands, Electronic Load Controllers are used. The ELC considered in this work consists of an uncontrolled rectifier and an insulated-gate bipolar transistor (IGBT) used as a chopper in series with a dump load.

The analysis and implementation of ELC using a Proportional Integral (PI) controller and Artificial Neural Network techniques have been done.

ANN work on the human brain's mechanism of problem-solving strategies. As reported in the literature [6-7] the Backpropagation algorithm of ANN has been successfully implemented to system controllers. This paper focuses on investigating the controlling performance of two variants of the Backpropagation algorithm, which are used to control the switching logic of the chopper. These techniques are namely Levenberg-Marquardt (LM) and Scaled Conjugate Gradient (SCG) algorithms.

II. SYSTEM DESCRIPTION

A schematic diagram of the SEIG-ELC system is shown in Figure.1 for supplying three-phase loads.

Figure.1. Schematic diagram of three-phase SEIG with ELC

The SEIG ELC system consists of a three-phase delta-connected induction generator driven by an uncontrolled Pico hydro turbine. The alternating voltage from SEIG is converted to dc by the three-phase rectifier circuit of ELC. The chopper switch is turned on by an appropriate gate driver circuit. The IGBT is switched on when the main load on SEIG is less than its rated value and switched off when the main load is at its rated value. The duty cycle of the chopper is controlled within 5-95% to provide safe operation of the chopper switch and enhance the system reliability [8]. After the chopper is switched on, current flows through the dump load and consumes the difference power (generated power-consumed power). The dump load or ballast load comprises of resistors used as heating elements. It is connected in parallel with the consumer load or the main load such that the total generated power is held constant.

PG=PC+PD (1)

Where PG is generated power of the generator, PC is the consumer load power, and PD is the dump load power.

III. DESIGN OF THREE-PHASE ELC AND APPLICATION OF ANN

the root mean square The uncontrolled rectifier and chopper switch will have the same voltage rating and it depends on the root mean square (rms) ac input voltage and the average value of output dc voltage. The dc voltage is calculated from [9] and [10]

Vdc= (3?2 VLL)/? = (1.35) VLL = 1.35 ? 230 = 310.6V (2)

where VLL is the (rms) value of the line-to-line voltage of SEIG. Considering an over-voltage of 10% of rated value during transient conditions, the peak voltage is calculated as

Vpeak=?2? (230 +10% ? 230) = 357.8V

The uncontrolled rectifier and chopper switch will have a current rating that is decided by the active component of input ac and calculated as

IAC = P/ (?3?V LL) = 2200 / (?3?230) = 5.52A (3)

where P is the power rating of SEIG.

Considering a distortion factor of the three-phase uncontrolled rectifier as 0.955 and crest factor of 2.0, the peak input current of ELC is calculated as

Ipeak = (IAC /0.955)?2=11.56A (4)

From these calculations, it is found that voltage and current ratings of the rectifier circuit and IGBT should be at least 400V and 11A respectively.

The rating of dump load resistance (RDL) is calculated by

RDL = (Vdc)2 / P = 43.85? (5)

Artificial neural networks (ANNs) can learn the approximate relationships between the inputs and the outputs of a system[11-14]. These are not restricted by the size and complexity of the system. The input and output training data for the controlling of gate pulse is taken from a polynomial function obtained by curve fitting of the PI controller. These controlling data are used to train the ANN controller by the Backpropagation algorithm.

The ANN controller produces optimized output which is compared with a saw-tooth carrier waveform to generate duty pulse using Pulse Width Modulation ( PWM) for IGBT to act.

This proposed controller is designed in neural network toolbox, implemented and simulated in MATLAB by the following two algorithms:

• A. Levenberg-Marquardt (LM) algorithm: The Levenberg-Marquardt Algorithm is used to provide a numerical solution in minimizing a complex nonlinear function. The performance function is the mean of the square of a nonlinear relationship.
• B. Scaled Conjugate Gradient (SCG) Algorithm: In Scaled Conjugate Gradient algorithms, a search is performed along the direction in which performance function decreases rapidly while minimizing the error.

Figure2 shows the ANN diagram. As shown in Figure 3 and Figure 4, it is observed that at each epoch the performance function is different for both the cases. As the gradient decreases and reaches close to zero, the performance function is minimized. This implies that the output becomes close to the target and the network is perfectly trained.

Figure2: ANN block diagram

12325352099310Time(sec)

Time(sec)

42125902097405Time(sec)

Time(sec)

28194001065530Voltage error(V)

Voltage error(V)

-1401581182687Voltage error(V)

Voltage error(V)

Figure 3: Error minimization by Scaled Figure 4: Error minimization by Levenberg Conjugate gradient Algorithm. Marquardt Algorithm.

IV. RESULTS AND DISCUSSIONS

The system comprising SEIG connected with a parallel connection of consumer load and dump load is considered. Simulation is carried out with and without controllers at very light load and loaded conditions. The results of Generator phase voltage (Va, Vb, Vc) and current (ia, ib, ic) of SEIG at load perturbation without ELC is shown in Figure 5. Figure 6 shows the Generator phase voltage and current of SEIG at load perturbation with ELC using a conventional PI controller.

-3038791561340Generator voltage (V)

Generator voltage (V)

-20288252167255 Time(sec)

Time(sec)

12579352171700Time(sec)

Time(sec)

78105-127000-5070471381124Generator voltage (V)

0Generator voltage (V)

-37719001094105Generator voltage (V)

00Generator voltage (V)

15093951710055 Time(sec)

Time(sec)

-19602451791657 Time(sec)

Time(sec)

250190516890Generator current (A)

Generator current (A)

-3118485410210Generator current (A)

Generator current (A)

lefttop00

Figure5: Generator phase voltage and current of SEIG Figure6: Generator phase voltage and current of without ELC at load conditions SEIG with ELC at load conditions

(2000?,300? at 2 sec,2000? at 3sec). (2000?,300? at 2 sec,2000? at 3sec).

From Figure5 it is observed that generator terminal voltage decreases on increasing the load at 2 seconds and increases on decreasing the load at 3 seconds. Whereas the waveforms of Figure 6 show that voltage and current are regulated to maintain constant compared to those without using a controller. The generator phase voltage (Va, Vb, Vc) and current (ia, ib, ic) of Self Excited Induction Generator controlled by ANN-based ELC using the algorithms Scaled Conjugate Gradient and Levenberg Marquardt are shown in Figure 7 and Figure 8 respectively.

-14986018548350014420851638935 Time(sec)

Time(sec)

-21628101638935 Time(sec)

Time(sec)

50801311150Generator voltage(V)

Generator voltage(V)

-3464560214630Generator voltage(V)

Generator voltage(V)

-438150-16192500

-19386551911350 Time(sec)

Time(sec)

12630151805432 Time(sec)

Time(sec)

19050358140Generator current (A)

Generator current (A)

-3101340576580Generator current (A)

Generator current (A)

Figure7: Generator phase voltage and current of SEIG

with SCG based ELC at load conditions Figure8: Generator phase voltage and current of (2000?, 300? at 2 sec, 2000? at 3 sec) SEIG 00with LM based ELC at load conditions

(2000?, 300? at 2sec,2000? at 3 sec).

Figure7 shows that the SCG algorithm-based control scheme gives better-controlled output voltage for the system. The waveforms of Figure 8 shows better-controlling efficiency with the LM algorithm compared to the other two controllers under varying load conditions.

The capacitor line current (ica), main load current(iMLa) per phase are shown in figures Figure9 and Figure10.

lefttop0

-3223531273050 Capacitor current (A)

Capacitor current (A)

-19077131758761 Time(sec)

Time(sec)

Figure9: Capacitor line current

-4445541655 Main load current (A)

Main load current (A)

13627102247018 Time(sec)

Time(sec)

Figure10: Main load current per phase

A comparison of the performance of LM and SCG algorithms working as controllers is represented in Table1.

Table1: Performances of both the algorithms.

Parameters Scaled Conjugate Gradient

algorithm Levenberg Marquardt algorithm

Time 1 sec 7sec

Epoch 63 iterations 29 iterations

Mean Square Error 0.136 0.0213

Regression value after training 0.994 0.999

Based on the results obtained from ANN implementation, it is observed that both the algorithms are comparable in terms of speed and accuracy. However, based on error minimization, the Levenberg Marquardt algorithm has better efficiency compared to the Scaled Conjugate Gradient. On the other hand, the Scaled Conjugate Gradient algorithm excelled in terms of speed (as found in average training iteration) on a simple Multi-Layer Perceptron structure (2 hidden layers).

V. CONCLUSION

A three-phase load supplied with a three-phase SEIG with an ELC is designed with appropriate ratings and modeled in MATLAB Simulink environment. An analysis of ELC using three control techniques namely PI controller, Levenberg-Marquardt and the Scaled Conjugate Gradient algorithms of artificial neural network controller has been presented in this paper. Simulated results show that the ANN controllers are more capable of maintaining constant voltage and power at varying loads compared to the conventional controllers. These can be applied in remote area power supplies effectively.

APPENDIX

The machine parameters are as follows: 2.2 kW, 3-phase, 4-pole, 50 Hz, 230 V, 7.8 A, delta connected,

Xls = 0.00446 k?, Xlr = 0.00446 k?, Rs = 0.00384k ?, Rr = 0.00288 k?

The magnetizing inductance (Lm) is related to the magnetizing current(Im) in the following manner:

Lm= 0.3177 for Im? 0.75

=0.3502 -0.0349 Im - 0.0017Im2 for 0.75 < Im ? 4.25

= 0.17667 for Im > 4.25

The parameters of ELC are -

Rdf=100000?, R1=10000?, Rf=15000?, Cf = 2.2?F.

REFERENCES

1. N. P. A Smith, Induction generators for stand-alone micro-hydro systems, IEEE International Conf. On Power Electronics Drives and Energy Systems, New Delhi, January 1996, pp. 669-673.
2. S. S. Murthy, O. P. Malik, and A. K. Tandon, Analysis of Self-excited Induction Generators, in Proc. IEEE, vol. 129, part C, no. 6, pp. 260-265, 1982.
3. A. K. Tandon, S. S. Murthy, and G. J. Berg, Steady-state analysis of capacitor self-excited induction generators, IEEE Trans. Power Apparatus and Systems, vol. PAS 103, pp. 612-618, Mar. 1984.
4. J. L. Bhattacharya and J. L. Woodward, Excitation balancing of a self-excited induction generator for maximum power output, in Proc. IEEE, vol. 135, pt. C, no. 2, pp. 88-97, 1988.
5. S.S. Murthy, B. Singh, S. Gupta, and B.M. Gulati, General steady-state analysis of three-phase self-excited induction generator feeding three-phase unbalanced load/single-phase load for stand-alone applications, in Proc. IEEE, vol. l50, no. 1, pp. 49-55, 2003.
6.  Jung-Wook Park, Venayagamoorthy, G.K., & Harley, R.G., (2005) MLP/RBF neural-networks based online global model identification of synchronous generator, IEEE Trans. Industrial Electronics, Vol. 52, No. 6, pp1685- 1695.
7. Venayagamoorthy, G.K., & Kalyani, R.P., (2005) Two separate continually online-trained neuro controllers for a unified power flow controller", IEEE Trans. Industry Applications, Vol. 41, No. 4, pp 906- 916.
8. S.N. Mahato, S.P.Singh, M.P. Sharma, Transient Performance of a three-phase Self-excited induction generator supplying single-phase load with electronic load controller, IEEE international conference,2010.
9. N. Mohan, T. M. Undeland, and. P. Robbins, Power Electronics: Converters, Applications, and Design, 2nd ed. New York: Wiley, 1995.
10. M. H. Rashid, Power Electronics, Circuits, Devices, and Applications,2nd ed. New Delhi, India: Prentice-Hall of India Private Limited,1996.
11. Halpin, S.M. & Burch, R.F.,(1997) Applicability of neural networks to industrial and commercial power systems: a tutorial overview, IEEE Trans. Industry Applications, Vol. 33, No. 5, pp1355- 1361.
12. Liu, Y., Cukic, B., & Gururajan, S. (2007). Validating neural network-based online adaptive systems: A case
13. study. Software Quality Journal, 15(3), 309-326.
14. MATLAB: Accelerating the pace of engineering and science. (2012). About MATLAB. Retrieved September 21, 2012, from
15. Sabahi, K. (2011, August). Application of ANN technique for DC- motor control by using FEL Approaches (pp.131-134). Paper presented at 5th International Conference on Genetic and Evolutionary Computing(ICGEC), IEEE.