How to dramatically reduce the amount of data M2M sensors transmit?
M2M sensors are predicted to generate more data than their human counterparts in the coming years. Unfortunately the price that will be paid for moving this traffic will be substantially lower than human data traffic. So it makes sense to think about ways to dramatically reduce the amount of data M2M sensors transmit.
How to do it?
In recent weeks I have been playing with RapidMiner. This program might be soon installed on a lot of Windows machines next to MS-Office. RapidMiner allows complete data mining layman to easily get hidden information out of the data they have at hand in files, Excels, Access, databases, etc.
RapidMiner shows how with some simple drag-and-drop in 5 minutes you can use complex algorithms like Neural Networks, Support Vectors Machines, Bayesian Classifiers, Decision Trees, Genetic Algorithms, etc. to make sense out of data.
The fact that you can easily train an algorithm to take a decision on your behalf could be a key factor to reduce the amount of M2M sensor data. So instead of sending all the data to a central point and making decisions there, you would put intelligence into the sensors.
This artificial sensor intelligence would not only be limited to single sensor failure. By applying genetic and swarm algorithms and copying mother nature, you would be able to have different sensors behave like for instance an ant colony. Individual sensors would start sharing alarm data and if enough or the right sensors agree then they would launch collective alerts.
Wireless technologies based on for instance White Spaces technologies can be used, and are already used for instance in Cambridge, to cheaply have many sensors communicate with one another. Also harvesting techniques should be used to avoid having to install batteries into the sensors.
The last part of the puzzle would be extra features in a M2M PaaS to manage the distribution of intelligence for de-centralized and self-organizing sensor networks. Sensors are likely to send data to a central server in which humans will have to train computers on what type of data is critical. Once the trained models are available, then they can be distributed to sensors. The M2M PaaS would focus from then on, on adjusting the algorithms in case certain alarms were not caught or when alarms were launched unnecessary.