-
Measuring Data Similarity for Efficient Federated Learning: A Feasibility Study
Authors:
Fernanda Famá,
Charalampos Kalalas,
Sandra Lagen,
Paolo Dini
Abstract:
In multiple federated learning schemes, a random subset of clients sends in each round their model updates to the server for aggregation. Although this client selection strategy aims to reduce communication overhead, it remains energy and computationally inefficient, especially when considering resource-constrained devices as clients. This is because conventional random client selection overlooks…
▽ More
In multiple federated learning schemes, a random subset of clients sends in each round their model updates to the server for aggregation. Although this client selection strategy aims to reduce communication overhead, it remains energy and computationally inefficient, especially when considering resource-constrained devices as clients. This is because conventional random client selection overlooks the content of exchanged information and falls short of providing a mechanism to reduce the transmission of semantically redundant data. To overcome this challenge, we propose clustering the clients with the aid of similarity metrics, where a single client from each of the formed clusters is selected in each round to participate in the federated training. To evaluate our approach, we perform an extensive feasibility study considering the use of nine statistical metrics in the clustering process. Simulation results reveal that, when considering a scenario with high data heterogeneity of clients, similarity-based clustering can reduce the number of required rounds compared to the baseline random client selection. In addition, energy consumption can be notably reduced from 23.93% to 41.61%, for those similarity metrics with an equivalent number of clients per round as the baseline random scheme.
△ Less
Submitted 12 March, 2024;
originally announced March 2024.
-
Optimizing a Digital Twin for Fault Diagnosis in Grid Connected Inverters -- A Bayesian Approach
Authors:
Pavol Mulinka,
Subham Sahoo,
Charalampos Kalalas,
Pedro H. J. Nardelli
Abstract:
In this paper, a hyperparameter tuning based Bayesian optimization of digital twins is carried out to diagnose various faults in grid connected inverters. As fault detection and diagnosis require very high precision, we channelize our efforts towards an online optimization of the digital twins, which, in turn, allows a flexible implementation with limited amount of data. As a result, the proposed…
▽ More
In this paper, a hyperparameter tuning based Bayesian optimization of digital twins is carried out to diagnose various faults in grid connected inverters. As fault detection and diagnosis require very high precision, we channelize our efforts towards an online optimization of the digital twins, which, in turn, allows a flexible implementation with limited amount of data. As a result, the proposed framework not only becomes a practical solution for model versioning and deployment of digital twins design with limited data, but also allows integration of deep learning tools to improve the hyperparameter tuning capabilities. For classification performance assessment, we consider different fault cases in virtual synchronous generator (VSG) controlled grid-forming converters and demonstrate the efficacy of our approach. Our research outcomes reveal the increased accuracy and fidelity levels achieved by our digital twin design, overcoming the shortcomings of traditional hyperparameter tuning methods.
△ Less
Submitted 7 December, 2022;
originally announced December 2022.
-
A Robust and Explainable Data-Driven Anomaly Detection Approach For Power Electronics
Authors:
Alexander Beattie,
Pavol Mulinka,
Subham Sahoo,
Ioannis T. Christou,
Charalampos Kalalas,
Daniel Gutierrez-Rojas,
Pedro H. J. Nardelli
Abstract:
Timely and accurate detection of anomalies in power electronics is becoming increasingly critical for maintaining complex production systems. Robust and explainable strategies help decrease system downtime and preempt or mitigate infrastructure cyberattacks. This work begins by explaining the types of uncertainty present in current datasets and machine learning algorithm outputs. Three techniques…
▽ More
Timely and accurate detection of anomalies in power electronics is becoming increasingly critical for maintaining complex production systems. Robust and explainable strategies help decrease system downtime and preempt or mitigate infrastructure cyberattacks. This work begins by explaining the types of uncertainty present in current datasets and machine learning algorithm outputs. Three techniques for combating these uncertainties are then introduced and analyzed. We further present two anomaly detection and classification approaches, namely the Matrix Profile algorithm and anomaly transformer, which are applied in the context of a power electronic converter dataset. Specifically, the Matrix Profile algorithm is shown to be well suited as a generalizable approach for detecting real-time anomalies in streaming time-series data. The STUMPY python library implementation of the iterative Matrix Profile is used for the creation of the detector. A series of custom filters is created and added to the detector to tune its sensitivity, recall, and detection accuracy. Our numerical results show that, with simple parameter tuning, the detector provides high accuracy and performance in a variety of fault scenarios.
△ Less
Submitted 23 September, 2022;
originally announced September 2022.
