Regarding the sensor network, the computer data-acquisition systems are based on small, low-power computers that communicate with each other via wireless links, although some base stations have more powerful computers (with greater power-supply requirements). One prototype system is a weather station near Eaglecrest Ski Area providing real-time weather and climate data for researchers at the University of Alaska Southeast and elsewhere (e.g., at the National Weather Service). A second prototype system, a small, low-power computer system, was deployed to the Lemon Glacier with geophones to monitor glacial lake drainage events. (It happened to record the magnitude 6.8 Queen Charlotte Island earthquake in 2004.) Wireless sensor network coverage will be extended to include the glacial monitoring program and to acquire data of specific interest to NOAA.
We propose grid computing [Foster et. al., 2001] for real-time data management in large sensor networks. Sensor capabilities are registered using description languages, and data are uploaded to the data warehouse via the data integration services to be available to applications via the data access services. Multivariate stream data mining will detect spatio-temporal patterns spanning multiple sensor streams. Since spatio-temporal relations across tuples of climate data are critical and the durations of events may vary, we depart from the traditional framework and label groups of adjacent tuples with one label. The training phase has two steps, first detecting “base patterns” in each stream using a frequent-subsequence mining algorithm. The next step extracts spatio-temporal patterns defining event types by connecting base patterns by temporal association. The runtime system builds a model of the ongoing event by identifying base patterns in the data streams and aligning them according to observed composite patterns.
Multiagent protocols can coordinate the wide range of sensors, analysis techniques, and subsystems required to identify complex, dynamic relationships in large volumes of data. Protocols more appropriate to this area than the contract net protocol (which we have used for vehicle health monitoring) will be investigated, and software components will be wrapped in agent interfaces (compliant with the FIPA reference architecture) to participate in the selected protocol. In light of the problems with testing, we propose the formal verification of a system’s design against its specification. For this, we have used the SPIN model checker [Holzman, 2003], and we propose to model-check critical aspects of the concurrency of the multiagent architectures developed. The grid infrastructure and appropriate services will be used as much as possible, and some coordinated activity of sensor networks will be automated.
[Foster et. al., 2001] Foster, C. Kesselman, S., Tueske, “The Anatomy of the Grid,” Intl. J. of Supercomputing App., 15(3), 2001.
[Holzman, 2003] Holzman, G., The SPIN Model Checker: Primer and Reference Manual, Boston: Addison-Wesley, 2003.