Context activity selection and scheduling in context-driven simulation

Abstract

Human activities in smart spaces are traced by sensors and logged, as sensor events, in the form of sensory values, when the sensors detect elements of the activities. The event-driven approach that models the combination of the sensor events is one of the most common human activity simulation approaches. However, this approach is scalewise challenged as activities and spaces get more complex. A large volume of sensor events demands more human efforts in modeling, and requires higher processing overhead. We observe that rather than simulating by combining sensor events, semantical abstraction could offer a scalable alternative in managing such complexity. In our previous work, we proposed a context-driven such an approach, which scales well in complex simulation. The approach evaluates current state space and advances the simulation loop by units of context, not by sensor events. By changing the domain of simulation from event to context, we could measure a remarkable performance advantage. Through the experiment, we noticed that activity design was critical to end performance. In this paper, therefore, we focus on modeling activities for a better fit to the contextdriven approach. We introduce a new activity model along with associated algorithms to select and schedule the activities. We also provide an evaluation of the performance and computational complexity of the algorithms.