Artificial neural networks for fault tollerance of an air-pressure sensor network
- Autori: Aronica, S.; Basilone, G.; Bonanno, A.; Fontana, I.; Genovese, S.; Giacalone, G.; Langiu, A.; Lo Bosco, G.; Mazzola, S.; Rizzo, R.
- Anno di pubblicazione: 2017
- Tipologia: Abstract in atti di convegno pubblicato in volume (Breve introduzione)
- OA Link: http://hdl.handle.net/10447/248861
A meteorological tsunami, commonly called Meteotsunami, is a tsunami-like wave originated by rapid changes in barometric pressure that involve the displacement of a body of water. This phenomenon is usually present in the sea cost area of Mazara del Vallo (Sicily, Italy), in particular in the internal part of the seaport canal, sometimes making local population at risk. The Institute for Coastal Marine Environment (IAMC) of the National Research Council in Italy (CNR) have already conducted several studies upon meteotsunami phenomenon. One of the project has regarded the creation of a sensors network composed by micro-barometric sensors, located in 4 different stations close to the seaport of Mazara del Vallo, for the purpose of studying meteotsunami phenomenon. Each station sends all the measurements to a collecting one that elaborates them with the purpose of identifying the direction and speed of pressure fronts. Unfortunately, four stations provide the minimum amount of data necessary to a reliable characterization of pressure fronts so that the failure of only one is a serious issue. Such failures regard blackouts, connection loss, hardware failures or maintenances. In this context we have developed a fault tolerance system that is based on neural networks. A feed forward neural network i is associated with each station i, and is trained to predict its measurements using as inputs the ones of the other three station j with j â i. In the normal condition, the collecting station receives the measurements from each station. In the case of failure of only one station k, the related neural network k can be used to predict the missing measurements. We have conducted preliminary experiments using a two layer feed forward neural network with sigmoid and linear activation functions for the hidden and output layer respectively. In order to simulate failures, we have removed group of data from each station measurements that follow inside a fixed temporal range. The related networks have been used to predict the missing measurements, and the mean square error (MSE) from the real measured value has been computed as performance index. The very low values of obtained MSE lead to the suggestion of a certain effectiveness of the proposed system.