Researchers Zesen Wang, Yonggang Li, and Lijuan Lan from the School of Internet of Things Engineering at Jiangnan University in China have developed a new approach to improve the reliability of soft sensing in industrial applications. Their work, published in the journal IEEE Transactions on Industrial Informatics, focuses on enhancing the performance and efficiency of Generative Adversarial Networks (GANs) to address the challenge of insufficient data in complex industrial scenarios.
Soft sensing is a technique used to infer hard-to-measure data through a large number of easily obtainable variables. However, in industrial settings, the availability of sufficient data can be a significant challenge, which can diminish the reliability of soft sensing. GANs have emerged as a promising solution for addressing this issue by generating synthetic data to supplement the available dataset. However, traditional GANs often fail to account for the mapping relationship between labels and features, which can limit their performance.
To overcome these limitations, the researchers proposed a multi-task learning-based regression GAN framework that integrates regression information into both the discriminator and generator. This approach allows the GAN to generate more diverse and realistic samples, thereby improving the quality of the generated data. Additionally, the researchers implemented a shallow sharing mechanism between the discriminator and regressor, which significantly enhances the algorithm’s operational efficiency.
The researchers also designed a dual data evaluation strategy that considers the importance of both training samples and generated samples. This strategy helps to increase the generalization of subsequent modeling, making the GAN more robust and adaptable to different industrial scenarios.
To validate the superiority of their method, the researchers tested it on four classic industrial soft sensing cases: wastewater treatment plants, surface water, CO2 absorption towers, and industrial gas turbines. The results demonstrated that the proposed approach significantly enhances the quality of generated samples while improving the algorithm’s operational efficiency.
The practical applications of this research for the energy sector are significant. For instance, in the case of industrial gas turbines, soft sensing can be used to monitor and optimize performance in real-time. However, the availability of sufficient data can be a challenge, particularly in remote or harsh environments. By using the proposed GAN framework, energy companies can generate synthetic data to supplement the available dataset, thereby improving the reliability and accuracy of soft sensing. This can lead to more efficient and cost-effective operations, as well as reduced downtime and maintenance costs.
In conclusion, the researchers have developed a novel approach to improve the reliability of soft sensing in industrial applications. Their multi-task learning-based regression GAN framework and dual data evaluation strategy offer a promising solution to the challenge of insufficient data in complex industrial scenarios. The practical applications of this research for the energy sector are significant, and the results demonstrate the potential of this approach to enhance the performance and efficiency of industrial operations.
This article is based on research available at arXiv.