Data Sampling Algorithm Based on Complexity-Entropy Plane for Smart Sensing Applications

Givanildo L. Nascimento

doi:10.1109/JSEN.2021.3116548

2021-11-15

PRE

Givanildo L. Nascimento *

Cristopher G. S. Freitas

Osvaldo A. Rosso

André L. L. Aquino

DOI：10.1109/JSEN.2021.3116548

原文链接

原文求助

摘要

En 中文

This work proposes a data sampling algorithm for smart cities applications based on sensor network infrastructure. Our algorithm identifies the sensor data behavior through the Causality Complexity-Entropy Plane and performs data reduction by removing redundant data without losing the system's properties. For this, we recognize the systems' dynamic changes in real-time through a delimiter, named Maximum Complexity Point (MCP). Thus, we determine when to update the sampling period to maximize the system's information content, i.e., the statistical complexity quantifier. To confirm the sampling adaptability, we apply our method in three different chaotic attractors: Rossler, Lorenz, and B-7. We compared our solution with two other sampling algorithms: (i) random histogram-based sampling and the L algorithm. We use the K-S test, the average Data Error, and the Causality Complexity-Entropy Plane to compare the results. Using our sampling approach, we observed K-S test distances less than 3% for chaotic maps and 1% for natural environments data. The best results were in the Data Error, showing an average error rate up to 13.4% lower when evaluating chaotic data and 15.7% lower when evaluating natural environments. Regarding the dispersion of points in the Causality Complexity-Entropy Plane, the sampled time-series reached regions of higher statistical complexity, indicating that they preserved information content, hence the original data's dynamics.

Keyword:

Information theory

data sampling

smart sensing application