Development of an islanding detection scheme based on combination of slantlet transform and ridgelet probabilistic neural network in distributed generation
One of the challenging issues for a grid-connected distributed generation is to find a suitable technique to detect an islanding problem. Islanding phenomena refers to the condition in which a distributed generation continues supplying power to a location at permissible voltage and frequency althoug...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2019
|
| Subjects: | |
| Online Access: | http://psasir.upm.edu.my/id/eprint/84182/1/FK%202019%2099.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | One of the challenging issues for a grid-connected distributed generation is to find a suitable technique to detect an islanding problem. Islanding phenomena refers to the condition in which a distributed generation continues supplying power to a location at permissible voltage and frequency although electrical grid power from the electrical utility is no longer present. Its drawbacks can lead to issues such as power quality, safety of utility personal, and even power generation protection. Thus, the technique must be able to differentiate islanding from other grid disturbances and disconnect distributed generation rapidly to prevent from mentioned problems. In this work, a new islanding detection technique is proposed based on combination of Slantlet Transform and Ridgelet Probabilistic Neural Network to detect islanding conditions from other disturbance for a 250-kW PV array connected to a typical North American distribution grid and a wind farm power generation system.
The proposed strategy includes several steps. In the first step, all possible detection parameters/ signals which can be affected by islanding occurrence in the system are measured locally for islanding and non-islanding conditions. Then, by means of the Slantlet Transform theory, the matrix data extraction for any decomposition level are computed and the best of them are selected as input data to feed to an effective classifier. Next, an advanced machine learning based on is utilized to predict islanding and none islanding states. In order to train Ridgelet probabilistic neural network, a modified differential evolution algorithm with new mutation phase, crossover process, and selection mechanism is introduced. Finally, the performance of the proposed methods is also assessed using the performance indicators such as various error measurement criteria, detection rate and false alarm and compared with recent works. Furthermore, to evaluate the efficiency of the proposed modified differential evolution for the training of ridgelet probabilistic neural network, four statistical search techniques, namely, particle swarm optimization, genetic algorithm, simulated angling, and classical differential evolution are used and their results are compared.
Results show that the proposed method is able to detect islanding conditions with 100% accuracy, 100% detection rate and 0% false alarm with detection time of less than 0.19s. Non detection zone decreases to around zero and the proposed method has the ability to detect islanding up to 0.1% power mismatch. The error measurements of the proposed method such as Mean Absolute Percentage Error, Mean Absolute Error, And Root Mean Square Error for islanding detection are less than 0.02% for ideal and noisy conditions which shows that the algorithm is not sensitive to noise. Based on the results the proposed method is accurate in detecting islanding phenomena and effective in noisy conditions. |
|---|