Intelligent task scheduling in sensor networks: introducing three new scheduling methodologies

title: Intelligent task scheduling in sensor networks: introducing three new scheduling methodologies
author: Wilbert L. van Norden
published in: April 2005
appeared as: Master of Science thesis
Man-Machine Interaction group
Delft University of Technology
thesis PDF (2.944 KB)
paper PDF (274 KB)


Ever more complex sensors have become available to maintain situational awareness during missions. Each has different capabilities and is therefore suited to one or more sensor functions. Choosing the best suited sensor for any sensor function is based on sensor capabilities as well as task attributes. In highly dynamic environments these characteristics can change rapidly, leading to a shift in sensor allocation. To increase performance of the entire sensor network the total set of sensors should be scheduled in a single system. This thesis puts forward and compares three new methods for scheduling prioritised tasks in a sensor network. The first scheduler is based on fuzzy Lyapunov synthesis. This scheduler uses a different buffer for each type of sensor function in which incoming task requests are placed. Whenever a sensor is available to execute a new task, a task is chosen from the largest buffer based on the sensors capabilities. This leads to a fast scheduling procedure with good performance. The second scheduler uses a genetic algorithm (GA) off-line to determine the optimal set of schedules for all sensors. Based on these optimal schedules a neural network (NN) is (re)trained to be used on-line. The use of the NN leads to a very fast scheduler, but one that guarantees neither optimality nor transparency. The third approach is a novel on-line use of a GA. This scheduler uses the GA to optimise the set of schedules, this time in a hybrid form. The solution of the scheduler currently used in sensor scheduling is included in the initial population; therefore the GA always improves the set of schedules. Using this implementation of a GA leads to a system that keeps optimising by re-scheduling. Tests showed that this novel on-line use of GA leads to a robust scheduler with highest performance.

blue line
University logo