Short-term electrical load prediction and related aspects

Abstract

This thesis examines a number of questions that arise in the process of forecasting and managing the load of an electrical power system, and presents some possible solutions. The study was based on the situation of one New Zealand Supply Authority, which is able to directly control the hot water heater component of its load. To the extent that the data worked with was obtained from this particular electrical system, the solutions found apply specifically to it. However the methods used to determine a model for the hot water heater load channels in response to switching, as well as the pulse filter and short-term load forecasting algorithms developed, are more generally applicable. The digital pulse filter algorithm is an improvement on the traditionally employed method of obtaining frequently updated readings of system load from kWh metering pulses. The hot water heater channel model that was found, enabled a reconstruction of uncontrolled load values from the measured controlled system load values to be undertaken. The ability of various short-term forecasting algorithms of the time series type to predict such load series was then examined. The different methods were designed to incorporate to various extents the features of the load and temperature series, and the effect of temperature on load. Comparisons of the methods' forecasting accuracies then pointed out those load features that it is most important to model in order to obtain better forecasts. (The results were of interest in that they showed that additional model sophistication did not necessarily imply more accurate forecasting algorithm performance.)