Power system congestion prediction using neural network algorithms

Abstract

As the number of electronic devices increases, there has become an increase of congestion on supporting power lines in a power system. This congestion, can increase the cost of power transmission due to the necessity of building new lines and can increase the cost of power in peak congestion time, depending on geographical location and transmission lines that are feeding the nearest hub. This rising power congestion also causes damage to power lines if they run too close to their limitations frequently. This thesis is exploring some of the available algorithms and tools that can be used to predict the congestion of a power system, which could then be coupled with a power system and monitored real time. This would enable the system management to make better decisions on future development/improvements to the system. The analysis methods presented are based upon a modified 24 bus IEEE Reliability Test System, and a pre-defined system load curve to simulate a real-time system. This load model is based upon the setup in the appendix of IEEE (RTS) [14]. This research led to finding a neural network algorithm approach to train an algorithm using past data and then used simulated real-time input to predict the congestion. The idea was for this prediction to be applied to a more dynamic system to see how well it would be able to find where the congestion is.