基于SVMD-FE和改进Informer的短期负荷预测  

Short Term Load Forecasting Based on SVMD-FE and Improved Informer

作　　者：印文博 谷卫 YIN Wenbo;GU Wei(Changzhou Railway Higher Vocational and Technical School,Changzhou 321000,China)

机构地区：[1]常州铁道高等职业技术学校,江苏常州321000

出　　处：《河北电力技术》2025年第1期51-58,共8页Hebei Electric Power

摘　　要：针对电力负荷非平稳性导致负荷预测精度低的问题,采用改进Informer模型对经过逐次变分模态分解(sequential variational mode decomposition, SVMD)算法和模糊熵(fuzzy entropy, FE)综合处理后的子序列进行预测,构建了SVMD-FE和改进Informer的预测模型。首先,采用SVMD算法对负荷数据进行分解,降低数据的非平稳性,利用FE算法对分解后的各子序列进行熵值重组;其次,在Informer模型中引入相对位置编码取代传统的绝对位置编码,以捕获序列数据内部的依赖关系,避免信息泄漏;再次,采用扩展因果卷积代替正则卷积来增加接收和增强局部信息提取。最后,结合某市负荷数据对比验证多种深度学习模型预测效果,结果表明该模型具有更高精度的短期负荷预测能力。In response to the problem of low accuracy in load forecasting caused by strong non-stationarity of power loads, an improved Informer model is adopted to predict the subsequences processed by the sequential variational mode decomposition(SVMD) algorithm and fuzzy entropy(FE).SVMD-FE and improved Infomer prediction models were constructed.Firstly, the SVMD algorithm is used to decompose the load data to reduce the non-stationarity of the data, and the FE algorithm is used to restructure the entropy values of the decomposed subsequences;Secondly, relative position encoding is introduced in the Informer model to replace the traditional absolute position encoding, in order to capture the internal dependencies of sequence data and avoid information leakage;In addition, extended causal convolution is used to increase reception and enhance the extraction of local information instead of regular convolution.Finally, by comparing and verifying the predictive performance of various deep learning models with the load data of a certain city, the results showed that the model has higher accuracy in short-term load forecasting.

关 键 词：短期负荷预测 相对位置编码 扩展因果卷积 变分模态分解 

分 类 号：TM715[电气工程—电力系统及自动化]