| Summary: | 碩士 === 國立成功大學 === 電機工程研究所 === 81 === A higher order statistics approach is attempted to
improve the accuracy of conventional ARMA (AutoRegression
Moving Average) model in short term power system load
forecasting. Due to the assumption of Gaussian process on the
time series of load, the model structure determination, the
parameter estimation and the model validation of conventional
approach are all based on the calculation of ACF (Auto
Correlation Function) and PACF (Partial Auto Correlation
Function), as in the well known Box and Jenkins method. In a
strict sense, the conventional approach should be confined to
the system of linearity, minimum phase and Gaussianity.
However, the problem of short term load forecasting does not
adhere to these limitations. The assumptions of the
conventional method on the short term load forecasting need to
be further investigated. This thesis employs higher order
statistics (i.e., cumulants) of the load time series
to extract maximum information contained in the data, and
thus to improve the accuracy of the short term load
forecasting. Since the higher order statistics approach can be
applicable to the problems of nonlinearity, non-Gaussianity,
nonminimum phase and colored noise, the power system load
studied in this thesis comprises different statistic patterns
of the whole system and the local load demands. The local load
demand includes those of the main transformer and the feeder
at a substation. A number of methods have been developed
to obtain the structure and parameters of the higher order
statistics model. Besides, the Monte Carlo simulation is
used to validate the method and computer programs developed.
|
|---|
