High-dimensional categorical data contains rich user-sensitive information, which poses huge privacy threats to users once leaked during data clustering and analysis. The existing K -mode method under local differenti...
High-dimensional categorical data contains rich user-sensitive information, which poses huge privacy threats to users once leaked during data clustering and analysis. The existing K -mode method under local differential privacy (LDP) always requires multiple user-server interactions, which not only has high communication overhead and computational cost but also makes user privacy vulnerable to malicious attacks during interactions. In this paper, we propose a non-interactive LDP K -mode clustering estimation method for high-dimensional categorical data. We first perform dimensionality reduction on each user data locally through the Fsketch algorithm. Then, we perturb the sketch data, ensuring the perturbation satisfies LDP. The perturbed data is then submitted to the server for K -mode clustering. Finally, on the server, we directly estimate the Hamming distance on the perturbed data to achieve K -mode clustering analysis. It is theoretically proven that our perturbation method satisfies LDP and is unbiased. Compared with state-of-the-art methods, our scheme more accurately estimates the Hamming distance between high-dimensional categorical data, reducing communication overhead due to only one interaction. Extensive experiments demonstrate the effectiveness of our method on four real high-dimensional category data sets, in which our scheme has a smaller normalized intra-cluster variance and larger purity index under the same privacy budget.
Distributed generation (DG) injects new vigor into the power system, but it sometimes brings unfavorable factors influencing the power quality of distribution network. In this paper, the Dynamic Voltage Restorer (DVR)...
详细信息
Distributed generation (DG) injects new vigor into the power system, but it sometimes brings unfavorable factors influencing the power quality of distribution network. In this paper, the Dynamic Voltage Restorer (DVR) is introduced to the distribution network to improve the power quality. In view of that the detection of voltage compensation is the key factor that influences the effect of voltage compensation, a voltage direct detection method based on mathematical morphology is proposed. Firstly, the positive-sequence voltage with harmonics can be extracted by the instantaneous sequence-component decomposer. And then the harmonics is filtered by mathematical morphology filter. Lastly, the control variable of voltage compensation needed by DVR can be calculated from the voltage direct-detecting algorithm. The simulation results show that this method has the characteristics of favorable accuracy, real-time performance and simplicity, and assures the voltage quality of load side after compensation by DVR.
暂无评论