Neural Network Harmonic Filter for Microgrid System

Mat Syai’in

, Dedy Kurniawan Setiawan

, Agus Muhammad Hatta

, Yuniar Farida

Shipbuilding Institute of Polytechnic Surabaya (SHIPS/PPNS), Surabaya Indonesia

University of Jember, Jember Indonesia

Institut Teknologi Sepuluh November (ITS), Surabaya Indonesia

State Islamic University Sunan Ampel, Surabaya Indonesia

Keywords: Active Filter, Neural Network, Total Harmonic Distortion, Microgrid

Abstract: Development of electrical technology, especially converters technology, has made significant change inthe

characteristics of electrical power systems. Moreover, the need of converting electrical signal from pure sine

to distorted sine from renewable energy based generators makes the use of converter technology increase.

Currently, the use of renewable energy based power plants together with traditional generators has become

commonplace, commonly known as microgrid systems. This is intended to improve efficiency, reduce

environmental pollution, and preserve nature. Microgrid systems have a very positive impact on the electric

power system. However, microgrids also cause negative impacts such as harmonics. This research is

developing active filter based on Neural Network concept. Neural Network is used as control strategies to

produce signals opposite harmonic signals. The simulation results show that the active filter based on neural

network can reduce the Total Harmonic Distortion (THD) in microgrid systems effectively.

1 INTRODUCTION

Development of electrical technology makes

significant changes in the characteristics of electrical

power systems. This is due to the dominance of the

use of digital technology in each sector. Digital

technology, especially converter technology, has

changed the nature of the electrical signal from pure

sine to distorted sine (Chakir, Kamwa et al. 2014;

Anwar, Elrayyah et al. 2015; Hashempour, Savaghebi

et al. 2016; Gonzatti, Ferreira et al. 2017). The more

affordability of renewable energy based generators

makes the use of converter technology increase.

Currently, the use of renewable energy based

power plants together with traditional generators has

become commonplace, commonly known as

microgrid systems (Abdelsalam, Massoud et al. 2011;

Li, Li et al. 2016; Cao, Zhang et al. 2018; Feng, Zeng

et al. 2018). This is intended to improve efficiency,

reduce environmental pollution, and preserve nature

(Dudurych, Rogers et al. 2012). Microgrid systems

have a very positive impact on the electric power

system. However, microgrids also cause negative

impacts such as harmonics. Harmonics can be

systemically detrimental if it is not well controlled

(Setiawan et al. 2015).

Regarding to minimizing the negative effects of

harmonics, many researchers have developed filter

technology, both passive filters and active filters.

Active and passive filters have their own advantages

and disadvantages. One of the advantages of passive

filters is relatively cheaper price, but passive filter

applications only work effectively under relatively

constant loading conditions, which are common in

industrial areas. As for the active filter, the price is

quite expensive, but it can be applied to various

loading patterns.

This research developed an active filter by

employing abc-dq transformation. The advantage of

using dq frame is the signal will be easier to control

because the value in dq frame is not influenced by

time, which is very different from value in abc frame

that changes by the time.

In this paper, Neural Network (NN) control was

examined to replace the PI controller for controlling

the active power filter. Simulation was performed in

MATLAB after training the neural network

(supervised learning) and it is shown that results are

acceptable and applicable in grid system. Compared

to PI controller, the NN controller can be less

complicated and less costly to implement in industrial

control applications (Setiawan et al. 2009).

344

Syai’in, M., Setiawan, D., Hatta, A. and Farida, Y.

Neural Network Harmonic Filter for Microgrid System.

DOI: 10.5220/0008906000002481

In Proceedings of the Built Environment, Science and Technology International Conference (BEST ICON 2018), pages 344-349

ISBN: 978-989-758-414-5

2 LITERATURE REVIEW

2.1 Modeling of Microgrid Systems

Microgrid systems in this paper are operated stand

alone, with three components of power plant. The

first is conventional power plant (diesel power plant),

the second is direct photovoltaic (PV) systems, and

the third is PV systems with battery.

2.1.2 Diesel Power Plant Model

Diesel power plant (DPP) model is generally

constructed by a diesel engine and synchronous

generator (SG). The complete dynamic model of DPP

requires modelling diesel engine with speed control

and SG with the system voltage control as well as

clutch between the SG and diesel engine as shown in

(Fig.1) [15-16].

Voltage

Regulator

Speed

Regulator

Diesel Motor Synchronous Generator

Controller

Figure 1: DPP model

2.1.2 Photovoltaic Systems Model

The electrical power generated and terminal voltage

of Photo Voltaic (PV) module depend on solar

radiation and ambient temperature. The equivalent

electrical circuit describing the solar cells array used

in the analysis is shown in (Fig.2).

Figure 2: PV equivalent circuit

2.1.3 Modeling of Battery Storage System

In this work, it was developed a model of the Ni-MH

electrochemical battery. The equivalent circuit of the

battery storage system is presented as in (Fig.3) [19]:

Figure 3: Electrical equivalent model of the battery

The simulation of this paper is aimed to show the

effectiveness of proposed method to reduce total

harmonic distortion (THD) referring to IEEE 519-

2014. The harmonics source simulated in this paper

was harmonics that were produced by rectifier. The

rectifier model can be explain as follows:

Figure 4: Load model

In this simulation, there were two kinds of load

model. The first model is linear load containing R, L

and C. The second model is non-linear load

containing power converters.

2.2 Modeling of Harmonic Filter

Filter used in this simulation is harmonic active filter.

There are two types of harmonic filter modeling. The

first one is harmonic filter modeling based on

Proportional Integral (PI) control. This type was used

as comparison method to verify proposed method

performance. The second modelling is harmonic filter

based on Neural Network (NN). This type is proposed

method.

Neural Network Harmonic Filter for Microgrid System

345

2.2.1 Harmonic filter based on PI controller

Figure 5: Model of Active Harmonic Filter

Figure 6: Model of Proportional Integral (PI)

controller for Active Harmonic Filter

2.2.2 Harmonic Filter Based on Neural

Network

Figure 7: Model of Neural Network (NN) Controller

for Active Harmonic Filter

2.3 Neural Network Training Process

Structure of Neural Network used in this simulation

has four inputs and two outputs. The inputs are load

voltages/currents in d-q frame and APF (active power

filter) voltages/currents in d-q frame. NN is

employing single hidden layer with 7 neurons for

modelling control strategy. Sigmoid function is used

as activation function. Sample data used as training

data is given in (Fig.8).

Figure 8: one of some pattern of training data

For convenience of data presentation, the data

training is provided in the graph to reduce number of

table rows due to the huge numbers of training data.

The upper box graph of Fig.9 is the input data and the

lower box graph of Fig.9 is the output. The

verification performance of proposed method is

conducted in the next section.

3 RESEARCH METHODS

This paper is aimed to simulate the reduction of THD

(total harmonics distortion) in Microgrid systems

using active filter that was designed in dq frame and

controlled using NN (Neural Network). The NN

controller that was designed in this paper had 4 inputs

and 2 outputs. The inputs were load signals in dq

frame and APF (active power filter) signal in dq

frame. The output was the controlled signals

representing source signals. This signal was drived to

pure signals. The complete design of simulation,

including power generation model, load model, and

filter model, is expresed in the flowchart at (Fig.9).

BEST ICON 2018 - Built Environment, Science and Technology International Conference 2018

346

START

Microgrid data

preparation

Modeling of Microgrid

component including

generator, load and filter

Simulation macrogrid system

with filter based on PI

controller

THD is within

the limit?

Taking the NN data training

Training process to get

Structure NN weight that use

as Filter Controller

Testing performance of NN

active filter for several cases

All THD is

within the limit?

Implementation in the Real

Microgrid Systems

STOP

Figure 9: Research Flow Chart

Several procedures simulated in this chapter are

aimed to verify the performance of active harmonic

filter based on Neural Network (NN). The

microgrid system used in this simulation is operated

stand alone (off grid) . The microgrid power plant is

combination of diesel power plant , PV and PV plus

battery. All the load models used in this simulation

were rectifier. Microgrid model used for simulation

is described in (Fig.10).

Figure 10: Model of Microgrid System

4 RESULTS AND DISCUSSION

There were two kinds of signal that were evaluated in

this simulation. The first was voltage signal and the

second was current signal.

4.1 Voltage signals simulation

The original signal voltage before control strategic

was applied. is shown in (Fig.11).

In this step, the load is modelling for producing high

level of Total Harmonic Distortion (THD). From

(Fig.11), it can be seen that the THD level was

31.06%. This simulation is aimed to reduce the THD

using PI controller in comparison to NN controller.

Figure 11a: The current signals before control action

is implemented.

Figure 11b: The THD current before control action is

implemented.

Neural Network Harmonic Filter for Microgrid System

347

Figure 12: Voltage signals resulted by PI controller.

The signals voltages in (Fig.11) were controlled

using PI controller. The model of PI controller can be

seen in (Fig.7). The result is provided in (Fig.12).

From the figure, it can be seen that the THD was

decreased from 31.06% to 20.16%. The THD level in

this case was still far from the IEEE 519 standard, but

from the result, it can be concluded that the PI

controller can reduce the THD to 33.16%.

The next simulation was controlling voltage

signals in (Fig.11) using NN controller. NN controller

model can be seen in Fig.8 and the result can be seen

in (Fig.13).

Figure 13: Voltage signals resulted by NN controller.

From the Figure, it can be seen that the THD

decreased to 19.62 %. This result has shown that

using NN controller can reduce THD better than by

using PI controller. By using NN controller, it can

reduce THD to 34.95 %. So, it can be concluded that

in this case, NN controller is better than PI controller.

4.2 Current Signals Simulation

Figure 14: The current signals before treated by

controller.

The current signals used as signals test in this

simulation can be seen in the (Fig.14). From the

figure, it can be seen that the signal had THD 5.38 %.

This signal was little bit over IEEE 519 standard. The

signal was then treated using PI controller. The signal

resulted by PI controller can be seen in the (Fig.15).

Figure 15: Current signals resulted by PI controller.

BEST ICON 2018 - Built Environment, Science and Technology International Conference 2018

348

Figure 15 gives information that THD current

decreased from 5.38% to 2.92 %. According to the

IEEE 519 standard, this signal was within the

standard.

Figure 16: The current signals resulted by NN

controller.

The current signals in Fig.14 were also controlled

by NN controller and the results can be seen in the

(Fig.16). From the figure, it can be seen that the THD

current decreased to 1.56 %. This simulation has also

proved that the NN controller gives better

performance than PI controller does. In this case, the

PI controller can decrease signal by 45.72 % while

NN controller can decrease THD level by 71%.

5 CONCLUSION

From the simulation results provided in section 3,

it can be concluded that the proposed methods, abc-

dq frame transformation and NN controller of Active

Harmonic Filter, have better performance in

comparison to PI controller of active Harmonic

Filter. In the voltage cases, the PI controller can

reduce THD by 33.16 % while NN controller can

reduce by 34.95 %. In other hand, in the current cases,

the PI controller can reduce THD current by 45.72%,

while the NN controller can reduce by 71%.

According to these data, the proposed method is

recomended as method for reducing THD, either

THD voltage or THD current.

REFERENCES

Abdelsalam, A. K., A. M. Massoud, et al. (2011).

"High-Performance Adaptive Perturb and

Observe MPPT Technique for Photovoltaic-

Based Microgrids." IEEE Transactions on

Power Electronics26(4): 1010-1021.

Anwar, S., A. Elrayyah, et al. (2015). "Efficient

Single-Phase Harmonics Elimination Method

for Microgrid Operations." IEEE Transactions

on Industry Applications51(4): 3394-3403.

Cao, X., J. Zhang, et al. (2018). "Joint Energy

Procurement and Demand Response Towards

Optimal Deployment of Renewables." IEEE

Journal of Selected Topics in Signal

Processing12(4): 657-672.

Chakir, M., I. Kamwa, et al. (2014). "Extended

C37.118.1 PMU Algorithms for Joint Tracking

of Fundamental and Harmonic Phasors in

Stressed Power Systems and Microgrids." IEEE

Transactions on Power Delivery29(3): 1465-

1480.

Dudurych, I. M., A. Rogers, et al. (2012). "Safety in

Numbers: Online Security Analysis of Power

Grids with High Wind Pentration." IEEE Power

and Energy Magazine10(2): 62-70.

Feng, J., B. Zeng, et al. (2018). "Evaluating Demand

Response Impacts on Capacity Credit of

Renewable Distributed Generation in Smart

Distribution Systems." IEEE Access6: 14307-

14317.

Gonzatti, R. B., S. C. Ferreira, et al. (2017). "Using

Smart Impedance to Transform High Impedance

Microgrid in a Quasi-Infinite Busbar." IEEE

Transactions on Smart Grid8(1): 428-436.

Hashempour, M. M., M. Savaghebi, et al. (2016). "A

Control Architecture to Coordinate Distributed

Generators and Active Power Filters Coexisting

in a Microgrid." IEEE Transactions on Smart

Grid7(5): 2325-2336.

Li, J., F. Li, et al. (2016). "S-shaped droop control

method with secondary frequency

characteristics for inverters in microgrid." IET

Generation, Transmission &

Distribution10(13): 3385-3392.

Setiawan, DK., M. Ashari, MH. Purnomo (2009),

"Diagonal Recurrent Neural Network Control of

Four-Leg Inverter for Hybrid Power System

Under Fluctuating Unbalanced Loads", Third

International Student Conference on Advanced

Science and Technology (ICAST)Seoul, Korea

Setiawan, DK., Y Megantara, BN Syah (2015),

"Three phase inverter of UPS control system for

harmonic compensator and power factor

correction using modified synchronous

reference frame" Electronics Symposium (IES)

PENS Surabaya, Indonesia

Neural Network Harmonic Filter for Microgrid System

349