Perform analytics at the edge - Internet of Things (IoT) Lens

Perform analytics at the edge

Data preparation is usually required before analytics can be performed at the edge.  There are various types of preparation and transformation that can be done at the edge to improve the sustainability of the solution:

  • Data filtering: IoT edge devices can filter out irrelevant data or noise from sensor data streams. For example, temperature sensors may monitor and report temperature at a high rate, but reflecting those changes as they come may not be useful to a specific application or use case. By filtering out the unnecessary data points, edge devices can reduce the amount of data that needs to be transmitted and processed, reducing communication related power and cloud processing costs.

  • Data aggregation: Devices can aggregate sensor data from multiple sources and over time to reduce the amount of data sent to the cloud. This can involve combining data from multiple sensors to create a single data stream, or aggregating data from multiple devices to provide a system-level view of performance or behavior.  This can reduce the number of connection requests as well as messages sent to the cloud.

  • Data enrichment: Sensor data can be enriched with additional contextual information, such as location or time data. This can enable more accurate analysis and insights, as well as provide additional context for cloud applications or systems. This reduces the need to enrich data in the cloud.  Local processing and data enrichment should be considered when the cost of doing the same operation in the cloud out-weighs the impact of larger payloads and additional local computation and resources.

  • Data normalization: Devices can normalize sensor data from different devices or sources to support consistency and compatibility with cloud applications or systems. This can involve converting data formats, units of measurement, or other data attributes to a common standard, reducing the need to do this in the cloud.

Devices can perform real-time analytics, predictive maintenance, anomaly detection, optimization, and security monitoring by analyzing sensor data and triggering alerts or actions based on predefined rules or machine learning models, without sending the source data to the cloud.

To perform analytics in an efficient and sustainable manner on the edge, use lightweight algorithms to improve performance and resource efficiency on IoT devices. Examples include algorithms such as decision trees or regression models that are less computationally intensive than deep learning models.  In addition, optimizing data structures can improve the performance and efficiency of analytics algorithms on IoT devices. Examples include data structures such as binary trees or hash tables that require less memory and processing power.