Predicting VOC Concentration Measurements: Cognitive Approach for Sensor Networks

Volatile organic compounds (VOCs) belong to special pollutants included in Kyoto protocol and in its updated versions. They are responsible for great and dangerous air pollution. The volatility of VOCs make them difficult to be computed and to be measured by means of appropriate sensors in terms of accuracy. Nowadays different methods have been adopted and approved by specific authorities. One of the most important is EPA25 issued by the American Environmental Protection Agency. As a matter of fact, EPA25 only works on set of complete data. In case of noncomplete set of data, we mean missing data issue due to different troubles, namely dysfunction concerning networks of sensors, thermal drifts, etc., EPA25 as well as other methods are not able to overcome the above issue that affects the prediction and the reliability. This research proposes an alternative and effective way, based on cognitive approach, to process VOC data delivered by a network of sensors by using a mixed genetic algorithm with an additional fuzzy-based procedure. A comparison with other techniques based on neural networks is also envisaged. The proposed modified genetic algorithm offers the best accuracy.