Neural network model-based training algorithm for transient signal analysis

The power quality disturbance analysis is becoming an essential issue because of widespread utilization of electronic nonlinear loads that have affected the operation of distributed power system network in residential and industrial areas. The quality of power system network plays an important role on the development and safety of power system industry. This paper proposes a wavelet network-based detection and location approach to power-quality disturbance analysis. The wavelet transform provides a suitable transient signal representation corresponding to a time-frequency plane which gives the related information relating to the analyzed signal. In order to extract power-quality disturbance features, the decomposition coefficient of wavelet transformation at each level is introduced and its mathematical calculation is established. The transformation detects and extracts disturbance features in the form of simultaneous time and frequency information and gradient or slope of the disturbance signal using the dyadic orthonormal wavelet transform. The processing phase contains a set of multiple artificial neural networks with wavelet transform coefficients as input signals. The simulation results and analysis indicate that the wavelet transform combining with neural network is sensitive to transient signal singularity detection.