Mid-long term power load forecasting based variable selection and transformer model(基于变量选择与Transformer模型的中长期电力负荷预测方法)

• Published in: Zhejiang da xue xue bao. Journal of Zhejiang University. Sciences edition. Li xue ban Vol. 51; no. 4; pp. 483 - 491
• Main Authors: 黄文琦(HUANG Wenqi), 梁凌宇(LIANG Lingyu), 王鑫(WANG Xin), 赵翔宇(ZHAO Xiangyu), 宗珂(ZONG Ke), 孙凌云(SUN Lingyun)
• Format: Journal Article
• Language: Chinese
• Published: Zhejiang University Press 01-07-2024
• Subjects: electricity time-series data(电力时序数据); mid-long term load forecasting(中长期负荷预测); multiple variable(多变量); transformer(transformer); variable selection(变量选择)
• ISSN: 1008-9497
• DOI: 10.3785/j.issn.1008-9497.2024.04.011

Accurate and effective load forecasting is very important for real-time operation and dispatching of power systems. In this paper, a prediction model that incorporates variable selection and sparse Transformer is proposed. Static and temporal variables are used as inputs to give full play to the information enhancement of static variables in the global time range. The variable weighting component is designed based on the gating mechanism with which different weights are assigned to the variables according to their relevance to the predicted output. A two-layer coding structure is designed for temporal feature extraction, attention is sparse, and future moment loads are predicted by multivariate inputs. The proposed model is validated using real power load data, and the experimental results show that it can improve the prediction accuracy and prediction efficiency of mid-long term load forecasting.(准确且有效的负荷预测对于电力系统的实时运行和调度非常重要。提出了一种融合变量选择与稀疏Transformer模型的预测方法，将静态变量和时序变量作为输入，充分发挥静态变量在全局时间范围内的信息增强作用，基于门控机制设计变量分权