engleski

Time Series Forecasting in Sensor Data Streams

We are witnessing a dramatic transition in our modern information-oriented world from the age of a single Internet-connected device per person into the age of the Internet of Things (IoT) where there is a multitude of connected devices per person. Many of these devices are sensors which generate a specific type of streaming data which is usually highly frequent, but mostly small in size. There are numerous examples of smart IoT applications where sensor data streams need to be automatically processed and analyzed by consumer components in real-time [1]. In this paper we focus on real-time fore- casting (i.e. prediction) of future values within sensor data streams. In this context, the data stream from a sensor may be considered as a time series that has to be forecasted. Sometimes, a sensor data stream can be highly correlated to several other sensor data streams (e.g. in the case of hydrological and weather sensors) which makes such a forecast quite challenging, as we present in the rest of this paper.

Internet of Things ; streaming data, time series ; forecasting

nije evidentirano