Measuring Complexity and Predictability of Time Series with Flexible Multiscale Entropy for Sensor Networks

Area of Science:

• Complex systems analysis
• Information theory
• Network science

Background:

• Time series complexity and predictability are crucial for sensor network topology and congestion control.
• Multiscale entropy (MSE) is a common method, but its component, sample entropy, has limitations with short time series.
• Existing methods like MSE and Composite Multiscale Entropy (CMSE) exhibit sudden changes in entropy values due to their similarity measurement.

Purpose of the Study:

• To introduce Flexible Multiscale Entropy (FMSE) as a more reliable and stable method for measuring time series complexity.
• To address the limitations of sample entropy in existing multiscale entropy methods, particularly for short time series.
• To enhance the performance of time series analysis for applications in sensor networks.

Main Methods:

• Developed Flexible Multiscale Entropy (FMSE) by introducing a novel similarity function.
• The new similarity function measures subsequence similarity with values ranging from zero to one, unlike the binary output of sample entropy.
• Evaluated FMSE on synthetic (white noise, 1/f noise) and real-world (vibration signals) time series data.

Main Results:

• FMSE demonstrates significant improvements in the reliability and stability of time series complexity measurement compared to MSE and CMSE.
• These improvements are particularly notable when analyzing short time series.
• FMSE provides more accurate complexity assessments, avoiding sudden changes in entropy values.

Conclusions:

• FMSE offers a more robust and stable approach to measuring time series complexity, especially for short datasets.
• The enhanced reliability of FMSE can lead to improved performance in sensor network applications, including topology and traffic congestion control.
• Flexible Multiscale Entropy represents a significant advancement in time series analysis techniques.