摘要
电网变电站安全运行是关系人民正常生活的重要工程。为提高汛期下电网运行的安全性,作者提出基于改进Transformer的电网防汛风险概率预测模型。首先根据数据特性选出Transformer作为时序数据预测的研究框架,研究位置编码对时序数据的影响,并提出时序数据位置自适应编码策略。然后针对逐点注意力机制计算方法的缺陷,提出局部数据感知增强策略;最后将两者结合,提出电网防汛风险预测模型。在变电站防汛数据集上的有效性实验结果表明,改进策略单一作用时有效,且共同作用时精度提高29.73%,实时性也没有下降。对比实验表明,基于改进Transformer的电网防汛风险概率预测算法优于其他多种深度学习时序预测算法,该模型在电网防汛预测方面表现出较好性能。
The safe operation of power grid substations is an important project that is related to the normal life of people.In order to improve the safety of power grid operation under the flood period,this paper proposes a probabilistic prediction model of power grid flood risk based on the improved Transformer.Firstly,Transformer is selected as the research framework for time-series data prediction according to the data characteristics,and the impact of location coding on timeseries data is studied,and the adaptive coding strategy for time-series data location is proposed.Then,the local data perception enhancement strategy is proposed to address the defects of the point-by-point attention mechanism calculation method;finally,the two are combined to propose the grid flood control risk prediction model.The effectiveness experimental results on the substation flood control data set show that the improved strategy is effective when it acts singly,and the accuracy is improved by 29.73%when it acts together,and the real-time performance is not degraded.The comparison experiments show that the algorithm in this paper outperforms many other deep learning temporal prediction algorithms and the model shows better performance in power grid flood control prediction.
作者
龙鑫玉
石英
林朝俊
LONG Xin-yu;SHI Ying;LIN Chao-jun(School of Automation,Wuhan University of Technology,Wuhan 430070,China)
出处
《武汉理工大学学报》
CAS
2022年第9期79-88,共10页
Journal of Wuhan University of Technology
基金
国家自然科学基金(52105528).
