Data Mining Techniques in Sensor Networks

Recent advances in pervasive computing and sensor technologies have significantly influenced the field of geosciences, by changing the type of dynamic environmental phenomena that can be detected, monitored, and reacted to. Another important aspect is the real-time data delivery of novel platforms. In this chapter, we describe the specific characteristics of sensor data and sensor networks. Furthermore, we identify the most promising streaming models, which can be embedded in intelligent sensor platforms and used to mine real-time data for a variety of analytical insights.

The management of massive amounts of geodata collected by sensor networks creates several challenges, including the real-time application of summarization techniques, which should allow the storage of this unbounded volume of georeferenced and timestamped data in a server with a limited memory for any future query. SUMATRA is a summarization technique, which accounts for spatial and temporal information of sensor data to produce the appropriate trade-off between size and accuracy of geodata summarization. It uses the count-based model to process the stream. In particular, it segments the stream into windows, computes summaries window-by-window, and stores these summaries in a database. The trend clusters are discovered as a summary of each window. They are clusters of georeferenced data, which vary according to a similar trend along the time horizon of the window. Signal compression techniques are also considered to derive a compact representation of these trends for storage in the database. The empirical analysis of trend clusters contributes to assess the summarization capability, the accuracy, and the efficiency of the trend cluster-based summarization schema in real applications. Finally, a stream cube, called geo-trend stream cube, is defined. It uses trends to aggregate a numeric measure, which is streamed by a sensor network and is organized around space and time dimensions. Space-time roll-up and drill-down operators allow the exploration of trends from a coarse-grained and inner-grained hierarchical view.

Ubiquitous sensor stations continuously measure several geophysical variables over large zones and long (potentially unbounded) periods of time. However, observations can cover neither every space location nor every time. Interpolation, i.e., the estimation of unknown data in each location or time of interest, can be used to supplement station records. Although in GIScience there has been a tendency to treat space and time separately, there is now great interest in analyzing data in both the domains. This suggests that integrating space and time would yield better results than treating them separately, when interpolating several geophysical fields. This chapter contributes to the investigation of spatiotemporal interpolators in a remote-sensing scenario. We describe two interpolation techniques, which use trend clusters to interpolate missing data. The former performs the estimation phase by using the Inverse Distance Weighting approach, while the latter uses Kriging. Both have been adapted to a sensor network scenario. The proposed techniques have been evaluated in a large air-climate sensor network. The empirical study compares the accuracy and efficiency of both techniques.

A PhotoVoltaic (PV) plant is a power station which converts sunlight energy into electric energy. In the last decade, PV plants have become ubiquitous in several countries of the European Union, due to a valuable policy of economic incentives (e.g., feed-in tariffs). Today, this ubiquity of PV plants has paved the way to the marketing of new smart systems, designed to monitor the energy production of a PV plant grid and supply intelligent services for customer and production applications. In this chapter, we start moving in this direction by fulfilling the urgent request of PV customers and PV companies to enjoy knowledge-based managing and monitoring services, integrated within a PV plant network. In particular, we illustrate a business intelligence solution developed to monitor the efficiency of the energy production of PV plants and a data mining solution for the fault diagnosis in PV plants.