| Summary: | 碩士 === 大同大學 === 電機工程研究所 === 85 === The premier task of power system planning is load forecasting. Budgeting, purchasing and constructing on schedule depend on an accurate and reasonable load forecasting. In the past, load forecasting focused on real power only, but not on reactive power. Power engineers almost predicted reactive power in the light of experience value of power factor, and counted the needed quantity of capacitors. In order to estimate the reactive power reasonably and accurately, this thesis will develop an applicable method to forecast reactive power for the basis of purchasing capacitors in distribution system planning.
In general, the variation of reactive power follows that of corresponding real power, and its quantity is affected by those factors, such as equipment, types of load, load composition, and so on. Each feeder is with different characteristics, so it''s not possible to adopt a formula to solve the problem for all substations. Therefore, under the consideration of easy use, less data collection and load representation of local distribution system, this thesis adopts summer peak daily records of feeders or main transformers in this year, to be trained in artificial neural network and then counts the increased amount of reactive power according to the growth or prediction of real power. It''s convenient that the daily records of DDCS system are selected as training data.
For peak-VAR prediction, this thesis adopts an unconventional method which uses the hourly incremental relationship of real power and reactive power. This method can fully represent the characteristics of reactive power for various load compositions at different time period. Besides, this method also can avoid the relationship distortion between real power and reactive power with the unknown on or off status of capacitors in distribution systems.
|
|---|
