BENCHMARKING OPEN-SOURCE IMPLEMENTATIONS FOR ENERGY TIME SERIES FEATURE EXTRACTION METHOD

The paper focuses on time series feature extraction technique benchmarking for consumer-side energy applications which can be used to build robust learning models for consumption forecasting and anomaly detection. More specifically we analyze various open-source implementations of the Matrix Profile algorithm for time series data mining available as software libraries written in the Python programming language. Several replicable benchmarking results are carried out on a reference large commercial building energy measurements data set while reporting aggregate run times in conjunction with the particularities of each algorithm. The work can serve as a practical guide for choosing appropriate algorithm implementations for new intelligent data-driven systems for smart building energy management.