AbstractShort-term load forecasting is important for reliable and economic operation of apower system. The aim of this research is the development of statistical models capable ofpredicting the short-term total system load for a small, isolated power system, utilising bothhistorical demand patterns and the underlying relationship between electrical demand andmeteorological conditions.
An essential prerequisite for the implementation of such a model is the continuousmonitoring of meteorological conditions. To enable such monitoring a data acquisition systemhas been developed and installed at a suitable location.
Initially, the value of forecasting was ascertained by considering the economicimplications for system operation in the absence of forecasting. Subsequently, the BoxJenkins time series approach was applied to the short-term load forecasting problem. Bothunivariate and multivariate (with weather inputs) models have been developed with areasonable degree of success.An off-peak load installation has also been monitored over an extended period andresults from this experiment support the use of off-peak load for emergency reservesubstitution. The practical aspects of using this low priority load for distributed load sheddinghave also been examined.
Although the time series techniques applied in this study are well proven, the maincontribution of this thesis has been the application of these statistical methods to the actual,local power system. Results indicate that by employing the multivariate technique an increasein accuracy over the univariate method was attainable. However, the improvement was notconsistent.Another departure was the use of the forecasts produced to investigate the effect offorecast accuracy on system operational costs. The univariate technique provided sufficientaccuracy (especially over 2h ahead) to approach optimum operation when used with aneconomic loading algorithm, assumming that an AGC system would make the necessary online adjustments to synchronised generation to account for forecasting errors.
|Date of Award||Dec 1992|
|Supervisor||D John Morrow (Supervisor) & Brendan Fox (Supervisor)|