面向电力工程信息处理的多元数据特征提取与融合  

Multivariate data feature extraction and fusion for power engineering information processing

作　　者：徐晓军 李奎 张秋琼 张方银 王曙光 XU Xiaojun;LI Kui;ZHANG Qiuqiong;ZHANG Fangyin;WANG Shuguang(Power Grid Planning Research Center,Guizhou Power Grid Co.,Ltd.,Guiyang 550002,China)

机构地区：[1]贵州电网有限责任公司电网规划研究中心,贵州贵阳550002

出　　处：《电子设计工程》2025年第5期120-124,共5页Electronic Design Engineering

基　　金：贵州省科技计划项目(202374GKJ7629)。

摘　　要：针对输变电工程影响因素多、数据样本复杂多变容易导致工程数据处理和预测难度增加的问题,文中提出了一种基于多元数据特征提取与融合技术的电力工程信息处理模型。该模型运用多个堆栈稀疏自编码器(SSAE)从电力工程各类数据中分别提取特征信息,借助竞争粒子群算法优化的回声状态网络(CSO-ESN)实现信息融合并输出预测结果。以电力工程信息中的静态投资数据为样本进行的多组对比实验结果表明,所提模型的预测误差范围为1.82%~5.95%,可以有效实现电力工程数据信息的处理与合理预测,具有良好的普适性与准确性。In response to the increasing difficulty of engineering data processing and prediction caused by multiple influencing factors and complex and variable data samples in power transmission and transformation engineering,this paper proposes a power engineering information processing model based on multivariate data feature extraction and fusion technology.The model uses multiple Stacked Sparse Auto-Encoder(SSAE)to extract feature information from various data of power engineering,and uses the by Competitive Swarm Optimized-Echo State Network(CSO-ESN)algorithm to achieve information fusion and output prediction results.The results of multiple comparative experiments using static investment data in power engineering information as samples show that the prediction error range of the proposed model is 1.82%~5.95%,which can effectively achieve the processing and reasonable prediction of power engineering data information and has good universality and accuracy.

关 键 词：电力工程数据 多元特征融合 堆栈稀疏自编码器 回声状态网络 趋势分析 

分 类 号：TN40[电子电信—微电子学与固体电子学] TP332[自动化与计算机技术—计算机系统结构]