Power Load Forecasting and the S&P500 Index Forecasting by Artificial Neural Network

• Main Authors: Chou, Yulin, 周郁霖
• Other Authors: Chuang, Shangjen
• Format: Others
• Language: zh-TW
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/77994855654234541137

碩士 === 國立高雄海洋科技大學 === 電訊工程研究所 === 101 === Electricity is an integral part of modern life, and it's also the driving force of economic development, whether the traditional industries or emerging high-tech industries made no sense! In recent years, Taiwan's industry and commerce become well-developed, and the service industry is booming, the power load is rising year after year. Therefore, this thesis presents the power load forecasting, to understand the power peak load accurate, and make sure the power company made the power reach fully utilized. In recent years, due to the global financial turmoil, the global economy has been effected and led to an economic downturn, so investment become an important concept of modern life, the stock investment finance and investment become the major choice for many people. Therefore, this thesis presents Standard& Poor's 500 index forecasting, to provide investors with a higher predictive accuracy for reference. In this thesis we use the back-propagation neural network and echo state network as the power load forecasting tools, the data were collected between January 1, 1992 to December 31, 1996, and has collected 1827 data in total, and take the previous day, two days before and seven days prior's the highest temperature, the lowest temperature, power loads, and the day's highest temperature, lowest temperature and the day of week as a data input variables. This thesis is training data between 1992-1994 and testing data between 1995-1996. S&P 500 Index use Back-propagation Neural Network and Radial Basis Function Network as the forecasting tools to predict the highest-price, lowest-price and close price on the day S&P 500 Index, making a comparison between BPN and RBF; the data were collected between January 4, 1965 to October 22, 2007, and has collected 10767 data in total, and take the daily open-price, a day before, and seven days prior's open-price, highest-price, lowest-price, the day of the week, and trading volume as a data input variables. The results found that the stock price growth rapidly in 1995-2007, the S & P 500 Index divided into two modes forecasting, one is training between 1965-1985 and testing between 1986-1994, the other is training between 1995-2000, testing between 2001-2007.