Statistical Modeling and Prediction of Power Consumption for Campus Buildings

• Main Authors: Ming-Wei Ye, 葉名瑋
• Other Authors: Sheng-Luen Chung
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/8842rz

碩士 === 國立臺灣科技大學 === 電機工程系 === 106 === Accurate predictions of human demand and resource models are important, such as: the response of power demand or the monitoring of the exiting ramp. We found that: the use of these energy and resources is closely related the pattern of people’s daily work. For example, the consumption of electricity in buildings is closely related to the user's daily routines-normal working day, and other exogenous factors such as climate. This paper studies the use of generalized additive models (GAM) to reactive power consumption model, and on this basis to predict. However, there is no systematic answer to the best GAM model. The goal of this thesis is to obtain the optimal model of the given dataset and the exogenous variables by using the systematic combination method of GAM formula. Once a mathematical model of these determinants is obtained, accurate predictions can be obtained. In addition, this paper uses Recursive least squares adaptive filter (RLS) and dynamic adjustment method to solve the problem of inaccurate prediction for realistic models. In the data presentation of the research methodology, we used the Campus’s data to study the power consumption. The results show that the use of (1) a year (2014) of historical power consumption data to forecast the next few years, and (2) only six months (January to June in 2014) of historical power consumption data to forecast the next six months. In the use of (1) a year of historical power consumption data to forecast the next few years, the mean absolute percentage error (MAPE) is best up to 8.48% and the coverage absolute error (CAE) is best up to 2.39%. In the use of (2) only six months of historical power consumption data to forecast the next six months, the MAPE is best up to 9.07%. The result of our research are better than those of the model formula used in the literature to predict the same power consumption data.