摘要
在处理电网资源池多类型负荷数据时,往往直接采用无迹卡尔曼滤波算法实现数据跟踪融合,忽略了多源异构数据采集的时间差异,导致负荷数据融合结果的相对误差较大。因此,提出新的多云异构电网资源池多类型负荷数据融合方法。针对电网负荷数据进行三次指数平滑处理,去除冗余数据信息。结合最小二乘法和拉格朗日插值方法,分别对预处理后的负荷数据时间序列进行采样周期调整、数据填充,完成数据配准处理,作为数据融合的基础。运用卷积自编码器、长短期记忆神经网络和挤压激励模块,构建具有约束对抗性能的多模态数据融合模型,以最小损失函数为目标对模型进行训练,通过训练后的模型可以实现负荷数据融合。实验结果表明,所提出方法负荷数据融合结果相对误差低于5%,满足了电网数据融合处理要求。
In the multi-type load data of power grid resource pool,the traceless Kalman filter algorithm is often directly used to achieve data tracking and fusion.Because it ignores the time difference of multi-source heterogeneous data acquisition,the relatively error of load data fusion result is large.Therefore,a new multi-cloud heterogeneous power grid resource pool multi-type load data fusion method is proposed.Triple exponential smoothing is performed on the load data of the power grid to remove redundant data information.Combined with the least square method and Lagrange interpolation method,the sampling period adjustment and data filling of the load data time series after preprocessing are carried out,respectively,and the data registration processing is completed as the basis of data fusion.Using convolutional autoencoder,long short-term memory neural network and squeeze excitation module,a multi-mode data fusion model with constraint confrontation performance is constructed.The model is trained with the minimum loss function as the goal.Load data fusion can be achieved through the trained model.The experimental results show that the relative error of the proposed method for load data fusion is less than 5%,which meets the requirements of power grid data fusion processing.
作者
刘永清
孟庆丰
陈又咏
蔡清远
程明
马朝晗
LIU Yongqing;MENG Qingfeng;CHEN Youyong;CAI Qingyuan;CHENG Ming;MA Chaohan(State Grid Information&Telecommunication Group Co.,Ltd.,Beijing 102211,China;State Grid Info-Telecom Great Power Science and Technology Co.,Ltd.,Fuzhou 350001,China;Fujian Yirong Information Technology Co.,Ltd.,Fuzhou 350001,China)
出处
《微型电脑应用》
2025年第12期49-53,共5页
Microcomputer Applications
基金
支撑电力数字空间构建的国家级自主可控研发基础平台研究与应用(52680021N00B)。
关键词
异构数据
电网资源池
电力负荷
对抗学习
数据融合
heterogeneous data
grid resource pool
power load
adversarial learning
data fusion
