Enhancing Reliability of Data-driven Soft Sensors for Industrial Processes

Strategic Insights -

Researchers at Zhejiang University have made significant contributions to the field of data-driven soft sensors, proposing two new models to improve their reliability and robustness in industrial chemical processes. The study, published in the journal Computers & Chemical Engineering, presents a solution to the issue of noise and outliers in historical process data, which can compromise the accuracy of soft sensors. The researchers developed Laplacian Huber regression and Laplacian piecewise linear regression models, incorporating a penalty term into the learning process to mitigate the effects of noise and outliers.

Key Takeaways:

  • The study highlights the limitations of traditional data-driven soft sensors in industrial processes, which are prone to noise and outliers in historical process data.
  • The proposed Laplacian Huber regression and Laplacian piecewise linear regression models offer a reliable solution to improve the accuracy of soft sensors.
  • The models incorporate a penalty term into the learning process, which helps to mitigate the effects of noise and outliers in historical data.
  • The study demonstrates the effectiveness of the proposed models in improving the reliability and robustness of soft sensors in industrial chemical processes.
  • The Laplacian Huber regression and Laplacian piecewise linear regression models have been validated through a simulation study and a case study in a real-world high-low transformer unit process.
  • The research was conducted by Xinmin Zhang, Ruikun Zhai, Zhihuan Song, and Manabu Kano at Zhejiang University, in collaboration with the State Key Laboratory of Industrial Control Technology.
  • The study's findings have significant implications for the development of reliable and robust data-driven soft sensors in industrial processes.

Statistics:

  • The study presents two new models: Laplacian Huber regression and Laplacian piecewise linear regression.
  • The proposed models have been validated through a simulation study and a case study in a real-world high-low transformer unit process.
  • The Laplacian Huber regression model outperformed traditional data-driven soft sensors in terms of accuracy and robustness.
  • The Laplacian piecewise linear regression model demonstrated improved accuracy and robustness compared to traditional soft sensors in the presence of noise and outliers.
  • The study was published in the journal Computers & Chemical Engineering in 2025.
  • The research was conducted by the Zhejiang University team, led by Xinmin Zhang.

Sources:

  • Zhang, X., et al. (2025). "Enhancing Reliability of Data-driven Soft Sensors With Stable Loss Function and Sample Graph." Computers & Chemical Engineering, 202. doi: 10.1016/j.compchemeng.2025.02.016
  • "Zhejiang University Develops Reliable Data-driven Soft Sensors." NewsRx. November 4, 2025.