Whole Time Series Data Streams Clustering: Dynamic Profiling of the Electricity Consumption

Area of Science:

• Computer Science
• Data Science
• Energy Systems

Background:

• Smart grid data is characterized by large volumes, high dimensionality, skewness, sparsity, and temporal fluctuations (daily, weekly).
• Sequential data streams exhibit evolving distributions, necessitating incremental processing with sliding windows due to memory constraints.
• Existing clustering techniques often focus on grouping observations within a single stream, rather than partitioning the entire data stream.

Purpose of the Study:

• To explore individual electricity usage characteristics from smart meter data.
• To develop and recommend suitable electricity tariffs for customers to reduce costs.
• To formulate real-time clusters based on smart meter data streams.

Main Methods:

• Investigation of various clustering algorithms and their enhancements.
• Application of incremental processing and sliding window techniques for time series data streams.
• Development of real-time clustering for smart meter data analysis.

Main Results:

• Identification of distinct electricity usage patterns through real-time clustering.
• Capability to formulate clusters in real time from high-volume smart meter data.
• Foundation for recommending personalized electricity tariffs based on usage behavior.

Conclusions:

• Real-time clustering of smart meter data is feasible and effective for understanding usage patterns.
• The developed algorithms enable dynamic tariff recommendations, benefiting consumers.
• This approach addresses the challenges of analyzing large, complex, and evolving time series data streams.