Forecasting Fluctuation in Power Supply

Electrical power systems are vulnerable to fluctuations in frequency due to variations that take place in demand and power generation capacity. In order to keep the frequency stable power generation needs to keep up with demand. A sudden increase in demand or a sudden drop in generation capacity causes an upward spike in frequency. The objective of the research is to detect precursor patterns in generation load, demand and other time series variables that lead to spikes in frequency.

This is a challenging problem as multiple, correlated data streams need to be mined concurrently. Furthermore, frequency spiking is a very rare event and robust methods are required to distinguish between genuine spiking signals caused by variation in dependent variables from noise that is constantly present in the stream.

We will be investigating the use of different pattern recognition methods including stream classification models and time series rule prediction methods.This project is undertaken by the DSRG in collaboration with a major power generation and transmission company based in New Zealand.