基于JIT-RVC算法的软测量的研究及其应用

JIT-Based soft sensors research and its alppication

针对传统JIT(Just-in-time)算法单纯以距离为数据选择的缺陷,以及局部模型鲁棒性差,稳定性不强的特点,文中一方面提出了与Jolliffe参数相结合的鲁棒最近相关算法,使得JIT算法能够在离群点出现的情况下正确的选择建模数据,另一方面,还将合成随机变系数概率模型算法(Ensemble RVC)作为JIT算法的局部模型构成了JIT-RVC算法,于此强化了传统JIT算法的非线性逼近能力和鲁棒性。最后,通过强烈干扰情况下的污水生化处理的BOD_5软测量预测仿真结果证明了算法的有效性,特别要指出的是相较于JIT-PLS算法,JIT-RVC算法的RMSE指标减少了55%,而相关系数提高了53%。

In order to address the problem of distarnce-based data selection method as well as the unrobustness and unreliability of local model in JIT algorithm, this paper proposed a robust nearest correlation algorithm by integrating correlation data selection method with Jolliffe parameters. This improvement makes the JIT algorithm best for data selection and insensitive to outliers. On the other hand, ensemble Randomly Varying Coefficient （RVC） model is incorporated into JIT algorithm as local model and then JIT-RVC algorithm is presented. By doing so, the nonlinearity ability and robustness ,3f the JIT algorithm are enhanced. The effectiveness of the algorithm is validated under intensive disturbance in BOD5 soft sensor predictio：a. Another point should be mentioned is that even if suffering from some outliers, the results show that RMSE and coefficient （r） are still improved by about 55% and 53% in comparison with convention JIT-PLS.