REACTIVE POWER FORECASTING FOR DISTRIBUTION SYSTEM BY USING AN ARTIFICIAL NEURAL NETWORK

• Main Authors: Li- Pei-De, 李培德
• Other Authors: Wen-Chen Chu
• Format: Others
• Language: zh-TW
• Published: 1998
• Online Access: http://ndltd.ncl.edu.tw/handle/41715988433557912775

碩士 === 大同工學院 === 電機工程研究所 === 86 === The premier task of power system planning is load forecasting. Accurate load forecasting can be effectively used not only to reduce cost but to maintain the security of system. In the past, load forecasting only focused on real power, but not on reactive power. Furthermore, the power industry of every country has moved toward liberalization. For Taiwan, reactive power demand will be regarded as an ancillary service if the vertically integrated power company is separated into three individual parts, generation, transmission and distribution. The supply of reactive power will no longer be free ; each reactive power service will charge. Hence, the reactive power forecasting has to be emphasized.Because every feeder has its unique characteristics, it is hard for us to use one equation to solve each substation problem. On the other hand, this thesis proposes a novel neural network approach- four-layer pruning and synthetic acceleration methods to overcome the slow learning speed of traditional neural network, and to solve the problem of neuron numbers. Besides, the DDCS records are utilized as the training data. In this way, the rapid and accurate forecasting objective can be reached.In this thesis, there are two types of reactive power forecasting. (1) the peak of reactive power, the hourly reactive power from off peak to peak load is forecast. (2) the hourly reactive power, the reactive power of twenty four hours a whole day is forecast. Both types are using increament relationship of real and reactive power to train the artificial neural network system. By doing this, the characteristics of reactive power for various load compositions at different time period can be fully responded, and the controversy phenomena of the unknown status of switching capactors can also be avoided.