Dstatcom Ai
Dstatcom Ai
Abstract— The decay of power quality is one of the princi- method in giving a better source current waveform. The step
pal issues of electrical distribution systems. One of the main size estimated using ANFIS learns faster and there is a lesser
reasons for this is the use of non-linear devices like the switch- static error in the estimation of the fundamental weights as
ing devices and the power electronics converters. Non-linearity compared to the other specified methods [7-8]. The com-
leads to detrimental effects on the utility of the power supply, parative results of ANN and ANFIS are analyzed and it is
system efficiency, reduction in power factor, etc. The reactive found that ANFIS gives a superior performance when com-
power is increased due to the reduction in power factor and it pared to ANN.
does not have any contribution in the transfer of energy, hence
its compensation is needed. In this paper, the Artificial Neural II. SYSTEM CONFIGURATION
Network (ANN) and Adaptive Neuro Fuzzy Inference System
(ANFIS) based control methods for a Three-Phase Distribution The circuit diagram in Fig. 1 is that of the DSTATCOM
Static Compensator (DSTATCOM) are proposed. This is used system. A reactive load has been connected to a three-phase
for compensation of the reactive power along with the current three-wire distribution system. The DSTATCOM is attached
related power quality issues that include harmonics. These with a DC bus through a Voltage Source Controller (VSC).
control methods are simulated in the MATLAB environment The VSC is realized with the help of three insulated-gate
with the help of the SIMULINK toolbox. The simulation re- bipolar transistors (IGBTs) having diodes which are con-
sults suitably demonstrate the performance of these methods nected antiparallel. The AC sides are connected to a point of
and it is found that source current waveforms with lesser dis- common coupling and loads. A Pulse Width Modulating
tortion are achieved with ANFIS as compared to ANN. controller has been used for providing pulses.
Keywords—Power Quality Compensation; MATLAB;
DSTATCOM; Synchronous Reference Frame Theory; Adaptive
Neuro Fuzzy Inference System; Artificial Neural Network;
I. INTRODUCTION
Power Quality remains one of the prime aspects of any
power distribution system. It has to follow IEEE 1547 stand-
ards throughout the different levels of the power system,
which are the Generation level, Transmission level, Distribu-
tion level, and Utilization level. Because of the extensive use
of automatic and power electronic based devices in an elec- Fig. 1. Fundamental Line diagram of the DSTATCOM system
tric distribution system, power quality problems have sub-
stantially increased. The distribution system has poor power III. CONTROL ALGORITHM
quality because of the inadequate reactive power during the For power quality compensation, compensating reactive
steady as well as the dynamic state. These problems gener- power is provided by the DSTATCOM depending on the
ally arise due to the presence of reactive and unbalanced needs of the load. The. DSTATCOM is controlled by several
loads [3,6]. A VSC DSTATCOM gives a robust and effec- algorithms. Here we make use a conventional algorithm to
tive methodology of providing shunt compensation in a line generate the data, and use two Artificial Intelligence based
that can help enhance the power quality [4-5]. Various con- control algorithms to enhance the results.
trol methods are available to derive the reference currents for
the DSTATCOM like the synchronous reference frame theo- A. SRFT Algorithm
ry (SRFT) [1]. The ANN and ANFIS based methods are able SRFT algorithm is also called the d-q based algorithm.
to determine and model the non-linear systems and can give This algorithm is used to produce pulses for the functioning
a better source current waveform and hence better power of the DSTATCOM. These pulses are provided at the ground
quality. terminal. Clarke Transformation is used for converting a-b-c
In this paper, the SRFT method is compared to ANN and to αβ0 reference frame and further αβ0 to d-q frame using
ANFIS based methods. The ANN based control method is Park Transformation.
efficient and viable as it is able to identify as well as model ݅ௗ
such non-linear systems [9]. However, the ANFIS based ܿݏˁ ݊݅ݏˁ ݅ఈ
ቈ݅ = ቂ ቃ ൨ (1)
method has more efficiency as compared to the ANN based െ݊݅ݏˁ ܿݏˁ ݅ఉ
Authorized licensed use limited to: SUNY AT STONY BROOK. Downloaded on August 13,2020 at 11:25:47 UTC from IEEE Xplore. Restrictions apply.
After getting the desired currents in d-q frame, quadra- 7.57E+00 -7.57E+00 -3.29E-04 1.24E-24
ture current is terminated and then direct current is added to -1.32E+01 7.52E+00 5.69E+00 4.87E-19
the error signal generated using the Proportional Integral (PI)
3.90E+00 -1.36E+01 9.74E+00 1.98E-15
control. This error signal is generated using 200V reference
and DC link voltage. Then d-q reference frame has been -2.55E-04 1.34E+01 -1.34E+01 7.94E-15
transformed back to the a-b-c frame by making quadrature 1.37E+01 1.32E-03 -1.37E+01 1.02E-13
current zero. This method is called Reverse Park Transfor- -4.31E+00 1.36E+01 -9.33E+00 1.72E-13
mation.
5.28E-01 -1.37E+01 1.32E+01 4.65E-13
݅ఈௗ ܿݏˁ ݊݅ݏˁ ݅ௗௗ -1.36E+01 -5.13E-04 1.36E+01 1.29E-09
ቈ݅ =ቂ ቃ ൨ (2)
ఉௗ െ݊݅ݏˁ ܿݏˁ ݅ௗ
Then the resulting current is compared with source cur- The performance of the network is analyzed by its mean
rent using hysteresis current controller in order to produce square error (MSE). Its equation is as follows:
Pulse Width Modulation (PWM) signals. These signals are ଵ 2
then fed to the main circuit. Emse = σே
ୀଵ ൣݔ െ ݔ ൧ (3)
ଶ
B. Artificial Neural Network (ANN) trained on data where ݔ represents the calculated output and ݔ de-
obtained by SRF Method notes the desired output and N signifies the number of train-
Artificial Neural Network is comprised of three layers: ing samples
Firstly, the input layer followed by the hidden layer and fi-
The best validation performance Mean Square Error
nally the output layer as shown in Fig. 2.
(MSE) obtained is 5.98251e-8.
Authorized licensed use limited to: SUNY AT STONY BROOK. Downloaded on August 13,2020 at 11:25:47 UTC from IEEE Xplore. Restrictions apply.
input layer that takes three inputs that are the three-phase
load currents and sends the input to layer 2 which performs
the process of fuzzification by making crisp values fuzzy.
Layer 3 defines the layer which matches output to input.
Layer 4 defines the normalization layer which is used to
measure the firing strength of normalization for a particular
given rule. Layer 5 defines the defuzzification layer which is
used for converting into crisp values from fuzzy values and
give the desired output direct current. Accuracy is bettered as
a result. Epochs used for training the ANFIS structure is 10
and the triangular membership functions has been used. The
number of nodes is 78, with 27 linear and 27 non-linear pa-
rameters. The number of training data pairs are 200001. 27
fuzzy rules have been used for the training. The training
MSE obtained is 0.000041.
(a)
(b)
(c)
Fig. 6. Optimized membership function for input variables (a) input1, (b) input2, and (c) input3
Authorized licensed use limited to: SUNY AT STONY BROOK. Downloaded on August 13,2020 at 11:25:47 UTC from IEEE Xplore. Restrictions apply.
Fig. 7. Training error for 10 epochs
(a)
(b)
Fig. 8. Control circuit diagram of (a) ANN algorithm and (b) ANFIS algorithm
IV. MATLAB SETUP OF DSTATCOM DSTATCOM model. Fig. 8 (a) - (b) shows the control circuit
ARRANGMENT diagram of the algorithms used. The following algorithms
are implemented using the MATLAB SIMULINK block set.
MATLAB Simulink model consists of a DSTATCOM
Simulation is accomplished in discrete mode at a maximum
system corresponding to the system arrangement presented
step size of 5e-06 with ode23s solver.
above in Fig. 1 comprising the source, load, DSTATCOM,
and control blocks. The first part of the circuit is basically a V. RESULT AND DISCUSSION
distribution system circuit including a combination of in-
ductance and resistance and a diode which makes the whole The behavior of the DSTATCOM using ANN and
system a nonlinear load. The above-mentioned SRF, ANN ANFIS control algorithms in enhancing the power quality of
and ANFIS based theories are used to simulate this
Authorized licensed use limited to: SUNY AT STONY BROOK. Downloaded on August 13,2020 at 11:25:47 UTC from IEEE Xplore. Restrictions apply.
the above distribution system can be observed by studying
̴ܸ௦ is the RMS voltage of the nth harmonic
the waveforms of the various parameters after compensation.
ܸ௨ௗ̴௦ is the RMS voltage of the fundamental frequen-
A. Results obtained using ANN based controller
cy
Fig. 9 shows the simulation results when the
Since THD is calculated using the amplitudes of the
DSTATCOM is controlled using ANN control algorithm and
harmonics, Fourier analysis is a method that helps determine
connected to the common coupling point. Here the source
THD. THD for source current and source voltage is calculat-
current is a sine wave approximately. In addition, the source
ed by using the FFT Analysis of the power GUI block. The
current and the voltage are in phase with each other at the
ANFIS control algorithm as compared to ANN control algo-
point. Hence, the source current is maintained at unity power
rithm provides better waveforms and lower THD. The THD
factor. Total Harmonic Distortion (THD) is found to be
graph for ANFIS shows more distortion as compared to the
1.65% for the source current.
one obtained using ANFIS.
B. Results obtained using ANFIS based controller
Fig. 10. Shows the results obtained on controlling the
DSTATCOM using ANFIS based technique. Here also the
source current and voltage at the common coupling point are
in phase with each other. Thus, unit power factor is obtained.
Total Harmonic Distortion of the source current as seen is
1.51% that is lesser than what is obtained using ANN.
ටσಮ మ
సమ ̴ೝೞ
THD = (4)
ೠ̴ೝೞ Fig. 11. THD Using ANN control algorithm
Authorized licensed use limited to: SUNY AT STONY BROOK. Downloaded on August 13,2020 at 11:25:47 UTC from IEEE Xplore. Restrictions apply.
TABLE III. SIMULATION PARAMETERS
Shunt Capacitor Capacitance and Initial Voltage 1500e-06 F, 156 V
Sampling Time 5e-06 s
PI Controller (Proportional, Integral) 0.4, 50
Source Inductance 2e-06 H
Load Inductance 0.003 H
Shunt Inductance 0.002 H
Source RMS Voltage 110 V
VI. Conclusion
In this paper, we have used Artificial Neural Network
(ANN) and Adaptive Neuro Fuzzy Inference System
Fig. 12. THD Using ANFIS control algorithm (ANFIS) control methods to improve the power quality
compensation. SRF theory, ANN as well as ANFIS based
TABLE II. COMPARISON RESULTS OF THD OBTAINED IN THE
DISTRIBUTION SYSTEM techniques have shown the satisfactory operation of
DSTATCOM. The ANFIS control technique has considera-
DSTATCOM Rise Peak Settling THD Conver- bly improved the performance of the DSTATCOM as com-
Control Time Overshoot Time gence
Algorithm Rate
pared to that of the ANN technique. The ANFIS based tech-
SRFT 0.001 4.55 0.0142 1.75% High nique utilizes both Fuzzy Logic as well as ANN algorithm
ANN 0.0013 4.42 0.0139 1.65% Low and provides higher accuracy and lower THD in source cur-
ANFIS 0.0014 4.02 0.0135 1.51% Low rent in conditions of varying load. MATLAB simulation
along with test results prove the efficacy of the mentioned
algorithms.
Thus, ANFIS has proven to be more efficient as well as
effective method in order to control the DSTATCOM for the
improvement of power quality. A considerable change in the
THD can be observed as ANFIS reduces the THD obtained
using ANN as mentioned above, and thus reduces the effect
of harmonics. This is how better results are obtained and the
power quality is improved.
REFERENCES
[1] Bhim Singh and Jitendra Solanki, ”A Comparison of Control
Algorithms for DSTATCOM,” IEEE TRANSACTIONS ON
INDUSTRIAL ELECTRONICS, VOL. 56, NO. 7, pp. 2738 – 2745,
JULY 2009.
[2] Alka Singh and Bhim Singh, “Power Quality Issues related to
Distributed Energy Source Integration to Utility Grids,”Annual IEEE
India Conference (INDICON), December 2010.
[3] R. Deepak Singh, T. Praveen Kumar, Dr.K.Sumanth,” Simulation of
SRF Based DSTATCOM With Grid Connected PV Generation
System Using Fuzzy Logic Controller For Reactive Power
Management,” International Journal of Advanced Research in
Electrical, Electronics and Instrumentation Engineering, Vol 5, NO.7,
pp 6493-6501,July 2016.
[4] H. Prasad and T. D. Sudhakar, “Power Quality Improvement by
Mitigation of Current Harmonics using D – STATCOM,” Third
International Conference on Science Technology Engineering &
Management (ICONSTEM), March 2017.
[5] Bhim Singh and Sabha Raj Arya, “Back-Propagation Control
Algorithm for Power Quality Improvement using DSTATCOM,”
IEEE Transactions on Industrial Electronics, vol. 61 ,No. 3 , pp. 1204
– 1212, March 2014.
[6] B. Singh, A. Chandra and K. Al-Haddad, Power Quality: Problems
and Mitigation Techniques, John Wiley and Sons, U.K., 2015.
[7] Ankita Arora and Prerna Gaur,”Comparison of ANN and ANFIS
based MPPT Controller for grid connected PV systems,” Annual
IEEE India Conference (INDICON), December 2015.
[8] Manoj Badoni, Alka Singh and Bhim Singh, “Adaptive Neuro Fuzzy
Inference System Least Mean Square Based Control Algorithm for
DSTATCOM,” IEEE Transactions on Industrial Informatics, vol. 12,
no. 2, pp. 483 – 492, April 2016.
[9] Yang Xiao-ping, Zhong Yan-ru and Wang Yan,” A Novel Control
Method for DSTATCOM Using Artificial Neural Network,”
Fig. 13. Comparison analysis of SRFT, ANN and ANFIS control algorithm CES/IEEE 5th International Power Electronics and Motion Control
on the basis of Settling Time, Peak Overshoot and Rise Time Conference, August 2006.
Authorized licensed use limited to: SUNY AT STONY BROOK. Downloaded on August 13,2020 at 11:25:47 UTC from IEEE Xplore. Restrictions apply.