基于差分隐私的个性化联邦电力负荷预测方案

A Personalized Federal Power Load Forecasting Scheme Based on Differential Privacy

为了实现兼具模型个性化和隐私保护个性化的电力负荷预测方案,文章提出了基于差分隐私的个性化联邦电力负荷预测方案。方案基于数据的缺失情况和时序特征进行集群式训练,得到适用于局部数据的本地个性化模型。在此基础上提出了个性化差分隐私保护方案,根据客户端到当前集群中心的距离调整隐私预算的分配,确保数据安全并实现客户端级别的隐私保护个性化。实验表明,算法在保证数据安全的同时,能训练得到效用较好的个性化模型。

In order to achieve a power load forecasting scheme with both model personalization and privacy-preserving personalization,this paper proposes a personalized federal power load forecasting scheme based on differential privacy.The scheme performs cluster-based training based on the missing cases and temporal features of data to obtain a local personalized model applicable to local data.On this basis,a personalized differential privacy protection scheme is proposed,which adjusts the allocation of the privacy budget according to the distance from the client to the current cluster center to ensure the data security and achieve the personalization of privacy protection at the client level.Experiments show that the algorithm can be trained to obtain a personalization model with better utility while ensuring data security.