Islanding Detection in Grid Connected Wind Plant

Using AI Technique
Srikanth Velpula, Ankathi Chandana, Ashok Valabhoju*
Kakatiya Institute of Technology & Science, Warangal-506015, India,
* ashokjntuk@gmail.com
Abstract.

This paper introduces an ingenious scheme of island detection in a dispersed

generation. The discovery of Islanding in distribution generation is one of the most
important aspects of a defensive aspect. Islanding is insensible for security of the system as
it may cause damage to personnel and equipment in the system due to the improper
islanding in the distributed generation system. Islanding detection is needed whenever
there is a sudden disconnection of wind power plant and fault. The system operates
through the current control mode, in normal or stable mode. The system switches to the
power control mode after sitting down. In the proposed work has made using the
MATLAB / Simulink software.

Keywords. Wind turbine (PV) System, Artificial Neural Network Method (ANN),
Voltage, Current, islanding, Levenberg–Marquardt (LM) algorithm and DWT–ANN.

1. INTRODUCTION
Modern day use of distribution production increases distribution production and worse
with benefits. Islanding is one such problem. The termination of a local distributed
generation (DG) system is termed as an islanding, unintended termination is termed as an
unplanned islanding. Islanding can be an error in the main distribution system, due to
power disturbance for a short period of time DG is cut off from the main grid.
There are many ways to find an island in a distributed generation system. Local and
remote routes practical and idle local routes. In terms of power generation and frequency it
is supplied to the PCC and compared to the threshold value while effectively the external
interference is injected into the PCC. The reversal of the operating path is an invisible
surface. It has low power quality.
Unintended islanding cause main problems which include:
a) Voltage & frequency within tolerable limits.
b) By supplying loads in DG system will risks to monitor staff safety, and
c) The closure of the DG system is out of phase which results in a rapid re-closure.
Therefore, in the most important energy system to find the occurrence of an islanding
condition.
The means of finding an island are broadly divided into 2 types namely remote and local.
Auxiliary local methods, distinguishing between active and passive:
1) Passive routes have a lot of undiscovered area (NDZ) and depend on the unattainable
condition to identify the island's population.
2) Effective methods have low NDZ, apart from that it will create a harmonic issue in
current.
In this paper a systematic literature review has been carried out thereby considering
previous research articles to understand the technical challenges. A recurrent neural
network-based islanding protection scheme by using generator speed change has been
reported [1]. A review of various islanding detection schemes for renewable based DG
system has reported in [2-4]. New ANN-based technique for islanding detection of DG
based power network was illustrated in [5] and Deep learning based hybrid scheme for
islanding detection in DG system is demonstrated in [6]. Islanding detection based on
ANN and novel S-transform for DG system has been explicated in [7] and an Islanding
detection method for Inverter-Based DG in Microgrid network has been reported in [8]. In
view of the above literature it has been concluded that the importance of identification of
islanding condition in a distributed power system network is very significant where grid-
connected wind power generation makes system more complex.

2. SYSTEM DESCRIPTION
In this paper, authors have taken the islanding inaction and the mistakes of non-
existing on the islanding situation as a research interest. A classification of islanding and
non-islanding conditions for islanding acquisition was made. Island and faults are given in
a grid connected wind plant at different parameters. The 3-phase voltage and current
signals are recorded from the side of the grid and PCC by changing the different
parameters. The data sets are simulated from the grid side on the PCC are supplied as an
input to ANN after reducing the undesirable noise from the signals. The algorithm used to
study the training process by ANN is the Levenberg Marquardt (LM) algorithm.

3. FAULT IN GRID-CONNECTED WIND PLANT

The short circuit fault has been applied at the grid side with a total of 50 fault cases
and recorded voltage & current signals on the grid bus (B120) and (wind plant). Fig.1-2
illustrations the voltages and current signals at bus B120 and B25 respectively.

The L-G fault, which is applied phase-A to ground, as shown in Fig.1. The three
phases of instantaneous voltage and current signals at the wind turbine bus (B25) are
changed, with the amplitude of voltage signals decreasing and the magnitude of current
signals increasing, as shown in Fig.2.
 Figure 1 3-phase signals at fault inception time 1.5s at Grid bus (B120):
(a) voltage waveform (b) current waveform,

Figure 2 3-phase signals at fault inception time-1.25s at Wind turbine bus (B25): (a)
voltage waveform (b) current waveform,

4. ISLANDING CONDITION IN A GRID-CONNECTED WIND PLANT

Grid disconnection is provided using a breaker and a total of 50 cases of Islanding
conditions are recorded on the grid bus and wind plant bus. Measure the voltage and
current signals on the side of the grid and at the wind plant. Fig.3-4 demonstrations the
voltage/current signals of the side grid B120 and wind plant bus B25 respectively.
Figure 3 3-phase signals at islanding inception time-1.5s at Grid bus (B120): (a) voltage
waveform (b) current waveform

Figure 4 3-phase signals at islanding inception time-1.5s at PV bus (B25): (a) voltage
waveform (b) current waveform

5. PROPOSED METHODOLOGY
The proposed MATLAB/Simulink model consists of a wind plant, grid and
existing power plants with different loads connected and a transmission cable.
Fig.5 shows a single line diagram of the connected wind plant when voltage and
currents are measured at PCC.
 Figure 5 Single line diagram of Grid-connected wind plant
It must be separated the electrical signals from the wind plant and the grid
on the PCC, and in the event of any deviation, they should be returned to ANN,
and identify the islanding status and issues a trip signal to the breaker to isolate the
wind plant to avoid damage with local loads caused by disturbance.

6. ANN FOR GRID-CONNECTED WIND PLANT

A multi-layered neural network is used for the island wind turbine detection. The
multi-layered neural network contains output, input and encryption layers that require
algorithm training to minimalize the error using the LM-algorithm. Discrete wavelet
transform is employed to eliminate noise. These databases are then processed through a
DWT-ANN processor to remove noise from the current 3-phase and voltage signals to read
the model as quickly and accurately as possible. Then, from all the voltage and current
signal on the Wind turbine bus and the Grid bus, we must take samples of two cycle data
(single-cycle or pre-islanding data, and one single-cycle data- fault or post-landing).
Finally, the standard deviation of the two-cycle data of current and voltage signals on
the grid bus and wind plant bus during fault, as well as islanding situation are used to
design input data set to train DWT-ANN. module, and provide the target output at '0 'if
incorrect, and '1' when connected to the ANN feed network to the ANN feed supply
provided 70 percent training, 15 percent testing, and 10 percent verification -15. Typical
data diversion of two three-phase current and voltage signals on bus120 and bus25 with
error modes and single-line circuit islands used to design a data set input training DWT-
ANN module. Various parameters have been used to generate input / output data set for
training and testing as noted in Table.1.
Parameters in particular start time, type of error and islanding method used,
number of data sets or cases of island residency, and error Grid bus and wind turbine we
used for training and testing using the ANN feed-forward network we have. is done
throughout the project to acquire an island Grid-connected wind turbine via ANN.
 Table.1 Parameters variation considered to generate the datasets for training & testing

7. MODELLING & SIMULATION

After that, we constructed the proposed ANN model with feedforward network
utilizing the Levenberg–Marquardt (LM) method with DWT-ANN processor to remove
noise from the signals, which we did using MATLAB Simulink. The Grid-connected wind
turbine system is depicted in Fig.6.
 Figure 7. MATLAB Simulink diagram of Grid-connected wind plant
ANN is trained using the MATLAB code, which includes parameters such as
performance goal, conversion, verification check, number of layers hidden in the
feedforward network, and number of epochs. Provide some parameter values in the code
and trained the network, check the MSE performance for each variation where the MSE is
the lowest, and then apply those values to the ANN Feed Transfer Network to find the
islanding condition to respond quickly and accurately. In the ANN feedforward network,
there is a single layer with six neurons, a single outgoing layer with six neurons, and a
hidden layer with a different number of neurons that are provided to the feed network in a
variety of conditions. ANN network. The ANN Feed-forward network training structure is
shown in Table.2, and the curves to ensure their performance during ANN between Epochs
and MSE are set out in Fig.7.
Table.2. Training architecture of different cases of Feed forward ANN

By looking at training curves, tests, and performance confirmation curves, we can

determine the best case. Case 2 is advanced because its three curves (training, testing,
and validation) will meet in one place that satisfies the island's living conditions
whenever a descent is possible in the system.

(a)

(b)
 (c)

(d)

Figure 7. Performance Validation curves during ANN using different size of architectures:
(a) case-1(6-20-6), (b) case-2 (6-25-6) (c) case-3 (6-30-6), (d) case-4 (6-35-6)

8. CONCLUSION
This paper is based on the detection of an islanding of wind plant using the ANN
method. Electrical power and current side grid and PCC signals at different inception
times are provided as input to the DWT-ANN module and are trained by the
Levenberg-Marquardt algorithm with 100 instances of ANN training data sets. Testing
/ training and verification is done to improve the performance of proposed method.

9. REFERENCES

[1]. Mehmet Bayrak, “Recurrent artificial Neural network-based island protection by

using generator speed deviation” (Sakarya University, Electrical-Electronics
 Engineering Department, Sakarya, Turkey.rences), Scientific Research and
Essay Vol.4 (4), pp. 212-216, April, 2009
[2]. Aziah Khamis, Hussain Shareef, ErdalBizkevelci, and Tamer Khatib, “A review
of islanding detection techniques for renewable distributed generation system” in
Renewable and Sustainable Energy Reviews, Volume 28, December 2013, Pages
483-493,
[3]. P. Mahat, Z. Chen and B. Bak-Jensen, "Review of islanding detection methods
for distributed generation," 2008 Third International Conference on Electric
Utility Deregulation and Restructuring and Power Technologies, 2008, pp. 2743-
2748,
[4]. R. S. Kunte and W. Gao, "Comparison and review of islanding detection
techniques for distributed energy resources," 2008 40th North American Power
Symposium, 2008, pp. 1-8.
[5]. V. L. Merlin, R. C. Santos, A. P. Grilo, J. C. M. Vieira, D. V. Coury, and M.
Oleskovicz, “A new artificial neural network based method for islanding
detection of distributed generators”, Int. J. Electr. Power Energy Syst., vol. 75,
pp. 139-151, Feb. 2016.
[6]. Xiangrui Kong, Xiaoyuan Xu, Zheng Yan,Sijie Chen, Huoming Yang, Dong
Han, “Deep learning hybrid method for islanding detection in distributed
generation”, in Applied Energy, Volume 210, 15 January 2018, Pages 776-785.
[7]. T. S. Menezes, D. V. Coury and R. A. S. Fernandes, “Islanding Detection Based
on Artificial Neural Network and S-transform for Distributed Generators”, 2019
IEEE Milan PowerTech, 2019, pp.1-6,
[8]. Karimi, M.; Farshad, M.; Hong, Q.; Laaksonen, H.; Kauhaniemi, K. “An
Islanding Detection Technique for Inverter-Based Distributed Generation in
Microgrids”. Energies, 2021, 14(1), 130.