Researchers from the University of Electronic Science and Technology of China, including Zexi Tan, Xiaopeng Luo, Yunlin Liu, and Yiqun Zhang, have developed a new method to improve the clustering of multivariate time-series data, which has significant implications for the energy sector. Their work, titled “Mask the Redundancy: Evolving Masking Representation Learning for Multivariate Time-Series Clustering,” was recently published in the IEEE Transactions on Knowledge and Data Engineering.
Multivariate time-series (MTS) clustering is a technique used to discover patterns in temporal data, such as the operation of machines or the output of solar power generation. However, these time-series often contain substantial redundancy, such as steady-state machine operation records or periods of zero solar power output. This redundancy can diminish the attention given to more discriminative timestamps, leading to performance bottlenecks in MTS clustering.
To address this issue, the researchers propose the Evolving-masked MTS Clustering (EMTC) method. This method is composed of two main modules: Importance-aware Variate-wise Masking (IVM) and Multi-Endogenous Views (MEV) representation learning. IVM adaptively guides the model to learn more discriminative representations for clustering, while MEV-based reconstruction and contrastive learning pathways enhance the generalization of the model.
The MEV reconstruction facilitates multi-perspective complementary learning, preventing the masking from premature convergence. Meanwhile, the clustering-guided contrastive learning facilitates the joint optimization of representation and clustering. The researchers conducted extensive experiments on 15 real benchmark datasets and found that EMTC outperforms eight state-of-the-art methods, achieving an average improvement of 4.85% over the strongest baselines.
For the energy sector, this research could lead to more efficient and accurate clustering of time-series data from various energy sources, such as solar, wind, and other renewable energy generation. This could help in better understanding and predicting energy output, optimizing energy storage, and improving overall energy management systems. The practical applications of this research could contribute to a more sustainable and efficient energy future.
The research was published in the IEEE Transactions on Knowledge and Data Engineering, a prestigious journal in the field of data engineering and knowledge management.
This article is based on research available at arXiv.