A decentralized approach for nonlinear prediction of time series data in sensor networks
EURASIP Journal on Wireless Communications and Networking - Special issue on theoretical and algorithmic foundations of wireless ad hoc and sensor networks
| Hi-index | 0.00 |
Sensor networks are a new class of distributed systems composed of a large number of densely deployed sensors, actuators, low power computation, and wireless communication devices. Sensor network research is still in its infancy. There is a large volume of exploratory research; however, few real systems are deployed and little experimental data from sensor networks is available to test proposed protocol design. Due to lack of experimental data and sophisticated models derived from such data, most data processing algorithms from the sensor network literature are evaluated with data generated from simple models. Through statistical performance analysis and algorithm evaluation using experimental data, we demonstrate the need to evaluate algorithms using realistic data corresponding to a wide range of parameter values and irregular topology input. Motivated out of a concern for data sensitivity and a lack of experimental data, we propose a scalable synthetic data generation framework to support systematic algorithm evaluation and robust algorithm design. We have two strategies to achieve scalable synthetic data generation. First, using case studies of concrete sensor network algorithms we identify a small number of parameters to manipulate in our synthetic data generation. This significantly reduces the search space from exponential to a manageable number. Second, we use empirical models derived from experimental data to guide simulation towards those portions of the space that represent real world scenarios. Guided by these two strategies, we implement algorithms to generate synthetic data either with similar characteristics as the experimental data, or exhibiting a wide range of data characteristics. In addition to the irregular topology data generation toolbox, we also provide ready-to-use test suites to encourage people to utilize more realistic data when evaluating their algorithms. Moreover, the sensor data replay addition to EmStar enhances EmStar simulation/emulation with more realistic sensor data traces.