Simulation study and experimental results for detection and classification of the transient capacitor inrush current using discrete wavelet transform and artificial intelligence

Abstract

This paper describes the combination of discrete wavelet transforms (DWT) and artificial intelligence (AI), which are efficient techniques to identify the type of inrush current, analyze the origin and possible cause on the capacitor bank switching. The experiment setup used to verify the proposed techniques can be detected and classified the transient inrush current from normal capacitor rated current. The discrete wavelet transforms are used to detect and classify the inrush current. Then, output from wavelet is acted as input of fuzzy inference system for discriminating the type of switching transient inrush current. The proposed technique shows enhanced performance with a discrimination accuracy of 90.57%. Both simulation study and experimental results are quite satisfactory with providing the high accuracy and reliability which can be developed and implemented into a numerical overcurrent (50/51) and unbalanced current (60C) protection relay for an application of shunt capacitor bank protection in the future.

An innovative harmonic‐based approach for inrush current detection in low‐loss transformers

“…As long as the training procedure is performed appropriately, these methods provide good accuracy, but the training process requires a lot of time and data [10,16]. System configuration change may require retraining, which could be time consuming.…”

“…These methods use time–frequency transformation techniques, such as wavelet and S transforms. They can provide higher reliability and accuracy compared to some other methods [8, 15, 16]. However, they have some drawbacks such as sensitivity to noise and disturbances along with imposing a higher computational burden and storage space [13, 15].…”

“…These methods use artificial intelligence algorithms, such as the artificial neural network, support vector machine, and so on. As long as the training procedure is performed appropriately, these methods provide good accuracy, but the training process requires a lot of time and data [10, 16]. System configuration change may require retraining, which could be time consuming.…”