ABSTRACT:
Artificial neural network (ANN) is
becoming an attractive estimation and regression technique in many control applications
due to its parallel computing nature and high learning capability. There has
been a lot of effort in employing the ANN in shunt active power filter (APF)
control applications. Adaptive Linear Neuron (ADALINE) and feed-forward multilayer
neural network (MNN) are the most commonly used ANN techniques to extract
fundamental and/or harmonic components present in the nonlinear currents. This
paper aims to provide an in-depth understanding on realizing ADALINE and
feed-forward MNN-based control algorithms for shunt APF. A step-by-step
procedure to implement these ANN-based techniques in MATLAB/Simulink environment
is provided. Furthermore, a detailed analysis on the performance, limitation,
and advantages of both methods is presented in the paper. The study is
supported by conducting both simulation and experimental validations.
KEYWORDS:
1. Adaptive
Linear Neuron (ADALINE)
2. Artificial neural network (ANN)
3. Feed-forward multilayer neural network (MNN)
4. Shunt active power filter (APF)
SOFTWARE: MATLAB/SIMULINK
CIRCUIT DIAGRAM:
Fig.
1. Shunt APF system configuration.
CONTROL SYSTEM:
Fig. 2. ADALINE used to extract the fundamental
active load current amplitude.
Fig.
3. Shunt APF control template using either MNN or ADALINE structures
SIMULATION RESULTS:
Fig.
4. Dynamic performance of the feed-forward MNN shunt APF for a trained
load scenario.
Fig.
5. Dynamic performance of the feed-forwardMNNshunt APF for untrained load
scenario.
Fig.
6. Dynamic performance of the ADALINE shunt APF.
CONCLUSION:
In this paper, two widely used ANN-based
shunt APF control strategies are investigated: 1) the ADALINE; and 2) the feed forward
MNN. A simple step-by-step procedure is provided to implement each method in
MATLAB/Simulink environment. The ADALINE is trained online by the LMS algorithm,
while the MNN is trained offline using the SCG back propagation algorithm to
extract the fundamental load active current magnitude. The performance of these
ANN-based shunt APF controllers is evaluated through detailed simulation and
experimental studies. Based on the study conducted in this paper, it is
observed that the ADALINE-based control technique performs better than the feed-forward
MNN. For untrained load scenario, the feed forward MNN fails to extract the
fundamental component, resulting in overcompensation from the dc-link PI
regulator. On contrary, the online adaptiveness of ADALINE makes it applicable to
any load condition.
REFERENCES
[1] P. Kanjiya, V. Khadkikar, and H. H.
Zeineldin, “A noniterative optimized algorithm for shunt active power filter
under distorted and unbalanced supply voltages,” IEEE Trans. Ind. Electron.,
vol. 60, no. 12, pp.5376–5390, Dec. 2013.
[2] B. Singh, K. Al-Haddad, and A.
Chandra, “A review of active filters for power quality improvement,” IEEE
Trans. Ind. Electron., vol. 46, no. 5, pp. 960–971, Oct. 1999.
[3] M. Popescu, A. Bitoleanu, and V.
Suru, “A DSP-based implementation of the p–q theory in active power filtering
under nonideal voltage conditions,” IEEE Trans. Ind. Informat., vol. 9, no. 2,
pp. 880–889, May 2013.
[4] V. Silva, J. G. Pinto, J. Cabral, J.
L. Afonso, and A. Tavares, “Real time digital control system for a single-phase
shunt active power filter,” in Proc. Conf. Rec. INDIN, 2012, pp. 869–874.
[5] A. Hamadi, S. Rahmani, and K.
Al-Haddad, “Digital control of a shunt hybrid power filter adopting a nonlinear
control approach,” IEEE Trans. Ind. Informat., vol. 9, no. 4, pp. 2092–2104,
Nov. 2013.
