Short-Term Load Forecasting Model Considering Multiple Time Scales

被引：0

作者：

Li, Dan ^{[1
]}

Tang, Jian ^{[1
]}

Zhen, Yawen ^{[2
]}

Zhang, Ke ^{[2
]}

机构：

[1] China Three Gorges Univ, Elect & New Energy Fac, Yichang 443002, Peoples R China

[2] Hubei Prov Key Lab Operat & Control Cascaded Hydr, Yichang 443002, Peoples R China

来源：

PROCEEDINGS OF 2023 INTERNATIONAL CONFERENCE ON WIRELESS POWER TRANSFER, VOL 3, ICWPT 2023 | 2024年 / 1160卷

关键词：

Short-term load forecasting; multi-time scale; convolutional neural network; MTSC-GRU prediction model;

D O I：

10.1007/978-981-97-0865-9_67

中图分类号：

TE [石油、天然气工业]; TK [能源与动力工程];

学科分类号：

0807 ; 0820 ;

摘要：

Most of the existing load forecasting methods are based on a single time scale for analysis and research, while the load power sequence has an apparent weekly cycle, daily cycle and seasonal time sequence characteristics. Therefore, this paper proposes an MTSC-GRU load forecasting model that considers the periodic characteristics of multiple time scales. First, the load sequence is divided into three different time scales, 15 min, daily and weekly, based on the temporal characteristics. Second, a convolutional neural network extracts temporal features from the sequences at three-time scales. Then a recurrent neural network is used to capture the long-term temporal dependencies in the load sequences, to learn the internal law of change of the loads, and to introduce the Dropout mechanism to avoid overfitting the model. Finally, the output side fuses and maps the GRU outputs at the three-time scales to realize the short-term day-ahead load timing prediction.

引用

页码：625 / 632

页数：8