On the Use of Stochastic Activity Networks for an Energy-Aware Simulation of Automatic Weather Stations

Introduction

Automatic Weather Stations (AWSs) (Sharan, 2014) are employed as sensing systems in extreme environments, such as desert, Antarctica (Reijmer, 2001), glaciers (Abbate et al., 2013) and deserts (Hobby et al., 2013). An AWS is normally composed of a number of sensors (humidity, wind speed, solar radiation, snow height) connected to a processing unit (CPU and memory). Since AWSs are usually far from mains power sources, they are often powered by batteries and exploit energy harvesting (solar panels or wind turbines, typically) to provide perpetual operation (Priya et al., 2009). AWSs can use communication interfaces, for example WiFi links, Radio bridges, GSM, or satellite links, to implement telemetry (Cesarini et al., 2013).

AWS deployments are built to last, in the order of years. Since after-deployment maintenance is generally difficult, expensive, or even impossible, designing an AWS is a critical task, the requirements being long-lived continuous operation, possibly in any working condition (Box et al., 2004). To address them, designers tend to over-provision energy related sub-systems, although this leads to higher costs and bigger size of the AWS.

Designers should be provided with computer aided design tools to study how hardware choices, e.g., battery and solar panel size, impact on the energy state and evolution of the system (Sabharwal et al., 2013). These class of tools, if used early during the design cycle, can help minimizing AWS design errors. Multiple approaches to analyse energy evolution of wireless systems have been proposed, in particular referred to Wireless Sensor Networks (WSN), which do share some features with AWSs. However, the following important differences hold:

• •

  AWSs are generally larger and more complex than WSN nodes.

• •

  The ratio between the energy stored in the battery and the energy spent during operation may be very different (up to two orders of magnitude) between AWSs and WSN nodes.

• •

  WSN nodes typically have a reduced set of sensors, while an AWS is normally equipped with a broad range of sensors.

• •

  Even though during the last 10/15 years a large number of research platforms for WSNs have been proposed, only few HW (telos, mica) and SW (TinyOS, Contiki) platforms are currently used in research, with an even lower industrial adoption. Instead, a wider and more heterogeneous panorama of industrial legacy platforms exists for AWSs (Cesarini et al., 2013), motivated by a broader range of different manufacturers, each one delivering it’s own hardware and software.

Considering the exposed differences, we believe that approaches focusing on WSNs cannot be easily adapted to effectively analyse AWSs. Thus, specific tools for AWSs are needed in order to design and install such systems with a higher degree of efficiency and accuracy.