基于对抗式域适应网络的气体传感器漂移补偿算法

Gas Sensor Drift Compensation Algorithm Based on Adversarial Domain Adaption Network

气体传感器漂移现象严重限制了其广泛应用。为降低漂移问题对气体传感器工作性能的影响,提出一种基于对抗式域适应网络的漂移补偿方法。该方法将域适应学习和对抗学习相结合,利用关联对齐(Correlation Alignment,CORAL)距离对齐源域和目标域的数据分布,从而使得源域样本训练的分类模型可以更好地在目标域样本上使用。利用该算法在公开数据集上进行漂移补偿实验,其识别气体的平均精度达到了85.9%,表明该方法可以有效补偿气体传感器的漂移,提高传感器的可靠性。

The drift phenomenon seriously restricts the performance of gas sensors in their various applications.In order to reduce the influence of drift problems on the working performance of gas sensor systems,a drift compensation method based on adversarial domain adaption network is proposed.The method combines domain adaption with adversarial learning,and aligns the source and target distributions with correlation alignment.Therefore,the trained classifier with labeled data can be better used with unlabeled target domain data.The algorithm proposed is used to perform drift compensation experiments on a public dataset,and the average accuracy of gas recognition can reach 85.9%,which means that this model can effectively compensate for gas sensors drift and improve the reliability of the gas sensor.