posted on 2024-11-11, 11:00authored byAli Yazdian Varjani
This thesis aimed to study all available short-term load forecasting methods in an attempt to suggest a solution (algorithm/structure) which gives the most appropriate forecast output for a typical input data set containing historical load data with or without weather variables input data. In this study, matters such as forecast accuracy, speed, development/implementation costs, and historical data validation were observed very closely. The two most successful time series based (conventional) load forecasting methods namely, auto regressive moving average and general exponential smoothing have been thoroughly studied and implemented, using available power utility input data. The relevant forecasting results along with discussion highlighting the inherent problems associated with conventional methods are presented. To fully examine the operation of a neural network for short-term load forecasting, a brief introduction to its theory and its implementation techniques is presented. Two major types of neural network structure (connection) for one-hour and 24-hour ahead load forecasting were selected. This is supplemented by two learning algorithms that are used for training and testing of the networks. To obtain the best possible forecast results, some structure modification was implemented along with the introduction of a modified learning algorithm. The sample forecast results for a standard and a proposed systems are shown. The proposed 24-hour ahead neural network forecast was compared with the results obtained from the most accurate conventional method using the same input data for training of the neural network and for model identification using this conventional method.
History
Year
1994
Thesis type
Masters thesis
Faculty/School
Department of Electrical and Computer Engineering
Language
English
Disclaimer
Unless otherwise indicated, the views expressed in this thesis are those of the author and do not necessarily represent the views of the University of Wollongong.