Abnormal power consumption data cleaning based on regular self-encoding and Optuna optimization

陈 慧; 陈 适; 郭银婷; 连淑婷; 王 康; 韦先灿

doi:10.3969/j.issn.1009-1831.2023.05.009

您当前的位置：

首页 >

文章列表页 >

Abnormal power consumption data cleaning based on regular self-encoding and Optuna optimization

更新时间：2026-03-05

- Abnormal power consumption data cleaning based on regular self-encoding and Optuna optimization
- Vol. 25, Issue 5, (2023)
- 作者机构：
  
  1. 国网福建省电力有限公司营销服务中心,福州,350001
  2. 福州大学电气工程与自动化学院,福州,350108
- 作者简介：
- 基金信息：
- DOI：10.3969/j.issn.1009-1831.2023.05.009
  CLC：
- Published：2023
- 稿件说明：
移动端阅览
陈慧, 陈适, 郭银婷, et al. Abnormal power consumption data cleaning based on regular self-encoding and Optuna optimization[J]. 2023, 25(5).
DOI：

陈慧, 陈适, 郭银婷, et al. Abnormal power consumption data cleaning based on regular self-encoding and Optuna optimization[J]. 2023, 25(5). DOI： 10.3969/j.issn.1009-1831.2023.05.009.

摘要

为有效解决用电信息采集系统中电量数据丢失问题

提出基于正则自编码器的缺失数据填补方法。首先

根据正则自编码器学习到的特征重构电量数据

实现缺失数据的修复。然后

通过对损失函数增加L21范数及正交约束实现正则化

提升模型的泛化能力

并采用Optuna实现超参数的自动寻优。最后

实际数据集的测试结果表明：与其他自编码器相比

正则自编码器能够较为准确地补齐缺失数据。

Abstract

In order to effectively solve the problem of consumption loss in the electric energy information acquisition system

a method of filling missing data based on regular self- encoders is proposed. Firstly

the energy data according to the characteristics learned by the regular autoencoder is reconstructed

and the repair of the missing data is realized. Then

regularization by adding the L21- norm is realized and orthogonal constraints to the loss function

the generalization ability of the model and uses Optuna to realize the automatic optimization of hyperparameters is improved. Finally

the test results of the actual data set show that compared with other autoencoders

the regular autoencoder can accurately fill in the missing data.

关键词

Keywords

references

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

No data

Related Author

No data

Related Institution

No data

AI问答

⁰