基于集成卷积神经网络的电网负荷数据分类  

Power grid load data classification based on integrated convolutional neural network

作　　者：朱正甲 ZHU Zhengjia(State Grid Jibei Electric Power Company Limited,Beijing 100054,China)

机构地区：[1]国网冀北电力有限公司,北京100054

出　　处：《电子设计工程》2024年第24期187-190,195,共5页Electronic Design Engineering

基　　金：2022年国网冀北电力发展部综合计划管理评价体系构建支撑服务项目(SGJB0000FCJS2250082)。

摘　　要：针对电力行业数据来源分散且维度较高,存在大量噪声,导致负荷数据分类非常困难的问题,提出了一种基于集成卷积神经网络的电网负荷数据分类方法。采用加权平均移动法生成负荷曲线,并降维处理电网负荷数据。构建双向激励函数,应用集成卷积神经网络训练降维数据集,区分正常数据和噪声数据。应用加权平均学习法通过构建新的网络对个体网络集成,建立集成学习后的数据集。利用不确定样本属性的验证、训练步骤,实现电网负荷数据分类。通过实验结果可知,该方法在电网负荷测量所属单元数值为0~1区间内的负荷数据分类结果为最优。Aiming at the problem that the data sources of the power industry are scattered and have high dimensions,and there is a lot of noise,which makes it very difficult to classify the load data,a power grid load data classification method based on integrated convolutional neural network is proposed.Use the weighted average moving method to generate load curves and reduce dimensionality to process power grid load data.The bidirectional activation function is constructed,and the integrated convolutional neural network is applied to train the reduced dimension data set to distinguish between normal data and noise data.The weighted average learning method is applied to integrate individual networks by constructing a new network,and the dataset after ensemble learning is established.Utilize the verification and training steps of uncertain sample attributes to achieve power grid load data classification.According to the experimental results,this method has the best classification results for load data within the 0~1 range of the unit to which the power grid load measurement belongs.

关 键 词：集成 卷积神经网络 电网负荷 数据分类 

分 类 号：TN391.1[电子电信—物理电子学]