P1.9.26 Exploitation of Gas Sensor Dynamics by the use of Echo State Networks

Hans-Ulrich Kobialka & Andreas Walte
Gas sensor arrays generate sensor data time series containing temporal information like gradients, noise patterns, and delays. We want to exploit temporal dynamics in gas sensor data for improving gas measurement in difficult environments, e.g. freight containers. For supervised training of pattern matching models we use Echo State Networks, a new approach for training large recurrent neural networks.