Data management and data analysis techniques for wireless sensor networks.

摘要

Fast developments in microelectronics and wireless technologies have made feasible the development of wireless sensor networks (WSN) composed of large numbers of small and smart sensors. Though cheap and small, the sensor devices have many resource limitations that introduce new challenges to the data management in sensor networks. Among all these limitations, energy is usually the primary concern when designing a sensor network algorithm. This dissertation presents two efficient approaches, data compression and data sampling, to minimize the energy consumption of the sensor networks. More specifically, it presents and evaluates an efficient data compression technique---the ALVQ (Adaptive Learning Vector Quantization) algorithm to compress the historical information in sensor networks with high accuracy. It shows how the ALVQ algorithm can be extended to compress multi-dimensional information and how the compressed data are transmitted in a sensor network while maximizing the precision. In addition, it proposes an efficient online sampling algorithm, Region Sampling, to retrieve a small fraction of the sensor data from the network while approximate the aggregate queries accurately. It demonstrates how a sensor network can be segmented into partitions of non-overlapping regions and how to use sampling energy cost rate and sampling statistics to compute the optimal sampling plan in different regions.; In addition to energy-efficient data transmission, rapid developments in flash memories have made possible the ability to store large amounts of data within individual sensors. This dissertation presents two index structures, MicroHash and MicroGF, for efficient retrieval of one dimensional and multi-dimensional sensor data stored in the flash memory of a sensor device. It shows how to exploit the asymmetric read/write and wear characteristics of flash memory in order to offer high performance indexing and searching capabilities. Finally it considers a typical sensor network application, distributed spatio-temporal similarity search , and shows how to combine local computations of lower and upper bounds in order to find the trajectories that are most similar to the query.