常英贤, 桂纲, 杨涛, 马广鹏, 邵晨, 汤泉. 基于区块链与安全计算的电能质量扰动分析方法[J]. 现代电力, 2023, 40(5): 732-741. DOI: 10.19725/j.cnki.1007-2322.2022.0097
引用本文: 常英贤, 桂纲, 杨涛, 马广鹏, 邵晨, 汤泉. 基于区块链与安全计算的电能质量扰动分析方法[J]. 现代电力, 2023, 40(5): 732-741. DOI: 10.19725/j.cnki.1007-2322.2022.0097
CHANG Yingxian, GUI Gang, YANG Tao, MA Guangpeng, SHAO Chen, TANG Quan. A Power Quality Disturbance Analysis Method Based on Blockchain and Secure Computation[J]. Modern Electric Power, 2023, 40(5): 732-741. DOI: 10.19725/j.cnki.1007-2322.2022.0097
Citation: CHANG Yingxian, GUI Gang, YANG Tao, MA Guangpeng, SHAO Chen, TANG Quan. A Power Quality Disturbance Analysis Method Based on Blockchain and Secure Computation[J]. Modern Electric Power, 2023, 40(5): 732-741. DOI: 10.19725/j.cnki.1007-2322.2022.0097

基于区块链与安全计算的电能质量扰动分析方法

A Power Quality Disturbance Analysis Method Based on Blockchain and Secure Computation

  • 摘要: 电力系统中的非线性特性极易引起电能质量扰动,破坏电网运行的稳定性。现有电能质量扰动分析方法大多将采集到的电信号传输到中心服务器进行统一存储,提取统计特征并利用机器学习构建模型。然而,在真实环境中存在隐私保护弱、设备环境复杂、模型过度依赖人工经验等问题。为此,提出了一种基于区块链与安全计算的电能质量扰动分析方法。首先,构建基于智能合约与联邦学习的私有链来保护数据隐私,通过无证书加密保障设备的身份可信,其次,利用基于Paillier密码体制的模型参数同态加密保护深度学习过程中的梯度安全,并建立基于变分模态分解与长短期记忆网络的电能质量扰动异常分析模型,以弥补传统统计特征建模的覆盖率与准确率不足。在真实搭建的微电网下的实验结果表明,该方法能够兼顾隐私性、可用性、安全性与精确性。

     

    Abstract: The nonlinear characteristics in power grid can extremely easy lead to power quality disturbance and destroy the stability of the power grid operation. Most of existing power quality disturbance analysis methods transmit the collected electrical signals to the central server to extract the statistical characteristic and then the models are built by machine learning. However, in real environment there are issues such as weak privacy protection, complicated facility environment and over-reliance on artificial experience in models. For this reason, a power quality disturbance analysis method based on blockchain and secure computation was proposed. Firstly, a private chain based on the smart contract and the federated learning was constructed to protect data privacy and by means of certificateless encryption the device identity creditability was ensured. Secondly, by use of the model parameters based on Paillier cryptosystem the gradient security during the deep learning process was protected by homomorphic encryption, and an anomaly analysis model of power quality disturbance(abbr. PQD) based on variational mode decomposition and long short-term memory network was established to cover the shortage of traditional statistical feature modeling in coverage rate and accuracy. The experimental results on a real microgrid show that using the proposed model, the privacy, usability, security and accuracy can be taken into account.

     

/

返回文章
返回