Abstract
For overhead long-distance high voltage direct current (HVDC) transmission lines, transients are produced due to complicated field conditions and lightning activities. To ensure reliable operation of protection devices, accurate recognition of faults and disturbances is quite critical. The most popular recognition methods include threshold-based ones which require the proper setting of the threshold value, and classifier-based ones that need suitable feature extractions. These methods depend heavily on the experience of engineers or experts and are ineffective in dealing with the variation of system parameters. In this paper, a transient recognition method based on stack auto-encoder (SAE) is proposed to characterize different transients and to avoid human interferences. A symmetrical ±500kv HVDC system is modeled to illustrate the performance of the proposed method. The effect of some factors, such as noises and conductors, are discussed and compared. The simulation results demonstrate that the proposed SAE-based recognition has excellent potential in transient recognition of practical HVDC systems.
Highlights
Considering some advantages, such as lower loss, more flexible control, and larger transmission capacity, high voltage direct current (HVDC) becomes a better choice for longdistance power transmission
As one of the most popular unsupervised learning methods, a Stacked Auto-Encoder (SAE) is a neural network consisting of multiple layers of sparse Autoencoders in which the outputs of each layer are wired to the inputs of the successive layer
The total recognition rate is 97.5% for lightning disturbance (LD) and lightning fault (LF), and 100% for ground faults (GF). It can be concluded the proposed stack auto-encoder (SAE)-based transient recognition is robust to the changes of the conductor, and it has great potential to be used in practical applications
Summary
Considering some advantages, such as lower loss, more flexible control, and larger transmission capacity, high voltage direct current (HVDC) becomes a better choice for longdistance power transmission. As one of the most popular unsupervised learning methods, a Stacked Auto-Encoder (SAE) is a neural network consisting of multiple layers of sparse Autoencoders in which the outputs of each layer are wired to the inputs of the successive layer It extracts features of signals layer by layer. As demonstrated by the symmetrical single polar HVDC shown in Fig., the singlepole-to-ground fault will result in a voltage increase of the normal pole. When a lightning discharge strikes on HVDC transmission systems, it can be modelled by a single current source before the insulation breaking-down. C. WAVEFORM ANALYSIS OF DIFFERENT TRANSIENTS From the above analysis, the transients on overhead HVDC transmissions are mainly responses of step voltage sources or impulsive current sources with different parameters, for example, magnitude, fault instant, fault location, and so on. The secondary code y2 of AE2 can be used as the features of raw input x
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
