DP-Denclue: An outlier detection algorithm with differential privacy preservation

As power systems evolve by integrating renewable energy sources, distributed generation, and electric vehicles, the complexity of managing these systems increases. With the increase in data accessibility and advancements in computational capabilities, clustering algorithms, including K-means, are becoming essential tools for researchers in analyzing, optimizing, and modernizing power systems. This paper presents a comprehensive review of over 440 articles published through 2022, emphasizing the application of K-means clustering, a widely recognized and frequently used algorithm, along with its alternative clustering methods within modern power systems. The main contributions of this study include a bibliometric analysis to understand the historical development and wide-ranging applications of K-means clustering in power systems. This research also thoroughly examines K-means, its various variants, potential limitations, and advantages. Furthermore, the study explores alternative clustering algorithms that can complete or substitute K-means. Some prominent examples include K-medoids, Time-series K-means, BIRCH, Bayesian clustering, HDBSCAN, CLIQUE, SPECTRAL, SOMs, TICC, and swarm-based methods, broadening the understanding and applications of clustering methodologies in modern power systems. The paper highlights the wide-ranging applications of these techniques, from load forecasting and fault detection to power quality analysis and system security assessment. Throughout the examination, it has been observed that the number of publications employing clustering algorithms within modern power systems is following an exponential upward trend. This emphasizes the necessity for professionals to understand various clustering methods, including their benefits and potential challenges, to incorporate the most suitable ones into their studies.INDEX TERMS Clustering algorithms, K-means clustering, power systems. FIGURE 3. K-means clustering and power system bibliometric trend per country.

Section: ) Dencluementioning

confidence: 99%

K-Means and Alternative Clustering Methods in Modern Power Systems

Miraftabzadeh,

Colombo,

Longo

et al. 2023

Improved Denclue Outlier Detection Algorithm With Differential Privacy and Attribute Fuzzy Priority Relation Ordering

Xia,

Chen,

Wang

et al. 2023

Outlier detection is an important method in data mining. Although Denclue algorithm is particularly good at finding clusters of arbitrary shape and detecting outliers, it does not protect the user's privacy well in the operation process. In this paper, differential privacy technology is introduced into Denclue algorithm to ensure the privacy security in the application of Denclue algorithm and outlier detection. Firstly, the differential privacy technology is used to add the Laplacian noise to the density to realize the sensitive information hiding between data objects. Secondly, in order to compensate for the decrease of outlier detection accuracy caused by noise, the information entropy weight distance was introduced to amplify the influence of important attributes in the algorithm, and the density function of entropy weight distance was used to calculate each data point. Finally, through the method of ordering fuzzy priority relation, a new measure index is defined by analogy to measure the degree of outliers between attributes. According to the measure index, the attributes are reordered and the weight distance of information entropy is improved. A differential privacy the Denclue outlier detection algorithm based on attribute fuzzy priority ordering (EAF-DP-Denclue) is proposed. The numerical results of the experiment show that the performance of EAF-DP-Denclue is more than that of traditional algorithms, and the identification process of EAF-DP-Denclue protects sensitive privacy information, and is better than that of the DP-DBScan outlier detection algorithm.

FedDBO: A Novel Federated Learning Approach for Communication Cost and Data Heterogeneity Using Dung Beetle Optimizer

Wang,

Chen,

et al. 2024

As an emerging distributed machine learning technology, federated learning has gained widespread attention due to its critical privacy protection mechanism. However, it also faces challenges such as high communication costs and heterogeneous client data.In order to address the above issues. This paper proposes a federated learning approach based on the dung beetle optimizer, named FedDBO. In this method, the model parameters uploaded from clients to the server are transformed into model scores. In each round of training, only some of the clients with high model scores need to be selected to upload their parameters to the server, thus reducing communication costs; Simultaneously, a model retraining strategy is introduced. After aggregating the model parameters sent by clients, the server performs a second iterative training on the aggregated model using its own metadata, thereby reducing data heterogeneity and improving model performance. In addition, a proof of convergence is provided, demonstrating that the model aggregated by FedDBO converges to the aggregated model of FedAvg after each training round. Finally, experiments indicate that when simulating various data heterogeneous environments on datasets, FedDBO exhibits higher accuracy and better stability compared to three other algorithms: FedAvg, FedShare, and FedPSO.