Power Quality Disturbance Eviction using SOM Neural Network
Swapnil B. Mohod1, Vikramsingh R. Parihar2*, Ketki R. Ingole3
1Department of Electrical Engineering, PRMCEAM, Amravati, India.
2Department of Electrical Engineering, PRMCEAM, Amravati, India.
3Department of Computer science and Engineering, Sipna College of Engineering and Technology, Amravati, India.
Power quality disturbances disintegrate the expected power waveforms and recurring disturbances leads to acute ramifications such as massive economic losses. Current researchers are indulged in addressing the issue of power quality disturbance problem as it is inherently a consumer-driven issue. In this paper, we have designed and optimized Artificial Neural Network (ANN) system to analyze the power quality disturbances and classify them with a higher degree of accuracy. The frequently observed power quality disturbances were predicted and classified using well-organized tools from ANN. Training of the Self Organizing Map (SOM) network was done using the different data partitioning methods and its performance on seen and unseen data was tested in terms of classification accuracy, Mean Square Error and correlation coefficient. The performance of proposed ANN system was verified with six types of power quality disturbances. By performing sensitivity analysis, numbers of inputs were reduced from 80 to 34(42.5%). Furthermore, observing that time elapsed per epoch per exemplary highly reduced from 2.150 ms to 0.49701µ-sec, and a number of connection weights also decrease from 2442 to 1892 (22.52%), which is reasonably good. Dimension reduction is also achieved by using the principle of sensitivity analysis which classifies the six types of power quality disturbances with a classification accuracy of 95.849%.
Key words: Kohonen self-organizing map, Power quality disturbance, Wavelet transform.