|
Data series resolution compression scale optimization based monthly electricity consumption forecasting |
DOI:10.19783/j.cnki.pspc.200539 |
Key Words:electricity forecasting resolution scale compression multi-step forecasting long short-term memory neural network |
Author Name | Affiliation | E-mail | WANG Fei | North China Electric Power University, Baoding 071003, China | | LI Zhenghui* | North China Electric Power University, Baoding 071003, China | zhenghuili@ncepu.edu.cn | LI Yu | State Grid Xinjiang Electric Power Co., Ltd., Urumqi 830018, China | | WANG Tieqiang | State Grid Hebei Electric Power Co., Ltd., Shijiazhuang 050021, China | | QIU Gang | State Grid Xinjiang Electric Power Co., Ltd., Urumqi 830018, China | | GUO Huaidong | North China Electric Power University, Baoding 071003, China | | MA Hui | Beijing Goldwind Smart Energy Technology Co., Ltd., Beijing 100176, China Xinjiang Goldwind Science Technology Co., Ltd., Urumqi 830026, China | | WANG Dongsheng | State Grid Jibei Electric Power Co., Ltd, Beijing 100054, China | |
|
Hits: 4386 |
Download times: 1459 |
Abstract:Accurate monthly electricity forecasting is the crucial basis to guide the operation plans arrangement of power grid corporation and guarantee the profitability of electricity retailers. To address the problem that improper selection of sample resolution during model training process based on artificial intelligence algorithms will seriously reduce the forecasting accuracy, a method for optimizing the compression scale of historical electricity data series resolution is proposed in this paper. First, the selection of data resolution compression scale is modeled as an optimization problem. Then, the data-driven method is used to solve the optimization problem. Finally, Long Short-Term Memory (LSTM) neural network algorithm is utilized to forecast the monthly electricity. As a result, a reasonable balance between data resolution and electricity forecasting step is achieved. The historical electricity data from PJM are utilized to verify the proposed method. The results show that the proposed method performs better than that without resolution compression scale selection, and the LSTM neural network combined with the proposed method shows the best forecasting performance. This work is supported by National Key Research and Development Program of China (No. 2018YFE0122200), Major Science and Technology Achievements Conversion Project of Hebei Province (No. 19012112Z), and Science and Technology Project of State Grid Corporation of China (No. SGHE0000KXJS1800163). |
View Full Text View/Add Comment Download reader |
|
|
|