传感器融合建模方法与PLSR理论拓展研究

Aimed to the multi-sensor fusion modeling, the principle and approach of a recursive partial least-squares (PLS) for on-line modeling is investigated based on the ordinary partial least-squares regression. The algorithm of simplified recursive PLS is proposed and the principles of the sliced-spline PLS as well as the inverse PLS are also studied in the project. A number of simulations and experiments have been conducted and the different algorithms are compared. The principle and algorithms above have been verified by using the experimental data and the simulation from the engineering application cases. The results demonstrate: 1) the algorithm of the simplified recursive PLS decreases the time-consuming in the computation greatly such that the algorithm is more suitable for the real-time application; 2) the algorithm of sliced-spline PLS can be implemented to the non-linear system modeling. The research on the principles and algorithms of the recursive PLS regression and sliced PLS regression extend to the applications of the multi-sensor data fusion with multi-dimensional and non-linear systems and on-line modeling requirement. Comparing with other modeling approaches (for example, artificial neural network, fuzzy logical method, etc.), extended PLS approach has an advantage in the model selection, the robustness, the explanatory ability, and less modeling time as well. The research of the extended PLS regression can provide a theoretical and practical modeling approach for the sensor signal modeling in the intelligent manufacturing systems and other complex systems.

根据传感器融合信号数据高维，非线性特点和建模要求，针对目前传感器融合建模方法中的问题和困难，研究探索分段样条偏最小二乘回归、逆偏最小二乘回归和递推偏最小二乘回归的原理与算法，将偏最小二乘回归拓展到高维非线性系统和实时应用领域，形成基于拓展的偏最小二乘回归的传感器融合建模理论，提供一种具有理论意义和应用价值的新建模工具。