Research on hardware assisted data cleaning algorithm for IoT Edge Computing

With the rapid development of Internet of Things(IoT) applications and edge computing, the data cleaning algorithm based on anomaly detection adopted by IoT terminals requires stream data processing, low power consumption and real-time performance. Based on the study of the core clustering feature (CF) tree data structure and processing window type of BIRCH stream clustering algorithm, this paper proposes an IoT terminal data cleaning algorithm supported by the IoT terminal Storage and Computation Integrated (SCI) architecture for the first time, and makes a low-power application design and comparative analysis on the Phase Change Random Access Memory (PCRAM) CF tree data storage and access. Prototype based experiments and results analysis shows that the proposed algorithm can meet the intermittent working mode requirements of IoT terminals, reduce power consumption, meet the real-time and online data cleaning performance requirements, and can be extended to support more complex machine learning enhanced data cleaning algorithms.