Soft-sensors have been used to estimate variables that are difficult to measure in real time. Many soft-sensor design methods have been proposed, however, the time-consuming trial and error such as parameter tuning is still necessary. Main reason is that the usefulness of each method have been often evaluated by a single dataset, and the robustness to other dataset and/or other process data was not evaluated. To solve this problem, this research proposes a robust parameter tuning method of locally weighted PLS (LW-PLS) model, based on the comparative study in twelve industrial processes such as the chemical, the pharmaceutical, and the food industries. The effectiveness of the proposed method is also validated by twelve industrial process data, and the results show that the proposed method can perform stable estimation considering nonlinearity while preventing over-fitting than the conventional methods.

软传感器已被用来估计难以实时测量的变量。已经提出了许多软传感器设计方法，但是，仍然需要耗时的反复试验，例如参数调整。主要原因是，通常只通过一个数据集评估每种方法的有效性，而未评估对其他数据集和/或其他过程数据的鲁棒性。为了解决这个问题，本研究基于化学，制药和食品工业等十二个工业过程的比较研究，提出了一种鲁棒的局部加权PLS（LW-PLS）模型参数调整方法。十二种工业过程数据也验证了该方法的有效性，