Wavelets and artificial neural networks in power system transient classification and short-term power load prediction

This thesis focuses on applications of wavelets and artificial neural networks in power system transients analysis, modelling, classification, and short-term power load prediction. A power system transient classification system framework based on wavelet transform preprocessing and probabilistic neural networks (PNN) is proposed in the thesis. A new type of neural networks, resource allocating networks (RAN), which can adjust its computing structure dynamically, is also investigated for the short-term power load prediction.; Wavelet modelling of power system transients is studied by examining (i) the capability of multiresolution analysis, (ii) time-scale representation of transient signals, and (iii) accurate detection and compact representation of transient signals, which are useful for power system transient recording, storing, and classification.; The PNN is used as a classifier in the proposed transient classification system. Experimental results show that the PNN has a great speed advantage in its training over backpropagation neural networks (BPN), which makes it a good candidate for real-time usage. The wavelet trans form preprocessing of the transient signals improved the performance of the PNN, and demonstrated the feasibility of the real-time transients recording and automatic classification.; The RAN is studied for short-term power load prediction because of the capability of RAN's adjusting its computing structure dynamically, which makes it a good candidate for modelling nonstationary signals. Experimental results on real data from Manitoba Hydro revealed that the short-term power load prediction is more accurate than other published results.