A Context-Driven IoT Middleware Architecture

The Internet of Things (IoT) refers to an environment of ubiquitous sensing and actuation from devices connected to the web backend. IoT applications leverage contextual information about entities in the system for reasoning and actuation. These context-aware applications are difficult to scale to the large amount of heterogeneous data in the IoT, as the current state-of-the-art is black-box, monolithic, application-specific implementations. We propose a middleware framework for context-aware applications that generates intermediate, reusable context extracted from input by breaking down applications into a set of functional units, or context engines. Leveraging existing IoT ontologies, we can replace application-specific implementations with a composition of context engines that use statistical learning to generate output, improving context reuse and reducing computational redundancy and complexity. We implement an IoT application using our framework, extracting residential user activity from plug loads, and demonstrate a reduction in computational complexity by 23% and execution overhead by 69%.