Noise Level Evaluation on Power Quality Disturbances Classification

Abstract

The detection of power quality disturbances is currently critical in the smart grid applications due to the bidirectional flow of power supply and demands at the main grid. The disturbance of the power quality may be occurred at utility or customer levels due to the unknown noises created by the parasitic components. Since the power data is in a time-series form, Long short-term memory (LSTM) networks are widely used due to its capability in recognizing temporal behavior from the inputs. This paper aims to study the performance of noise level susceptibility using LSTM network. The noise is inserted to the synthetic PQDs in term of Signal-to-Noise ratio with random noises generated from uniform white Gaussian. These noisy data is used to train and test the performance of LSTM model to understand the noise level interruption as LSTM model is popular to determine useful feature in the temporal domain. This work shows the sensitivity of the LSTM model towards unknown noises that is not seen during training phase. The complexity of harvesting all type of PQDs integrated with noises might be increased in the real-time PQDs classification.