基于最优定界椭球-极限学习机算法自适应软测量建模的应用

Application of Adaptive Soft Sensor Model Based on Optimal Bounding Ellipsoid-Extreme Learning Machine

针对传统滑动窗更新模型时忽略最新数据和待测样本相似性,以及即时学习未考虑相似样本和待测样本的时间间隔问题,采用基于最优定界椭球-极限学习机算法(optimal bounding ellipsoid-extreme learning machine,OBE-ELM)的自适应软测量建模方法将即时学习和滑动窗模型相结合来解决上述问题。首先用初始窗口数据建立ELM模型。当有待测样本到来时,利用SPE和T^2统计量判断修正模型的必要性;需要修正时,采用即时学习在最新窗口中寻找与待测样本相似的样本集并通过OBE动态修正ELM模型;否则用原有ELM模型直接预测输出。该方法的有效性通过合成数据集和连续搅拌反应釜仿真数据得以验证。

Owing to the traditional sliding window ignore the similarity of the latest data and the tested sample when model is updated,and just-in-time learning is not consider the time interval between the similar sample and the measured sample so a soft senor model based on optimal bounding ellipsoid-extreme learning machine(OBE-ELM)is proposed to solve the above problem.Firstly,the extreme learning machine(ELM)model is established by using the initial window data.When there is a sample to be tested,the necessity of revising the model is judged by using the SPE and the T 2 statistics.When the correction is needed,using just-in-time learning to look for the sample set in the latest window which similar to the tested sample and dynamically correct ELM model via OBE,otherwise the output is directly predicted using the original ELM model.The effectiveness of this method is verified by a synthetic data set and continuous stirred tank simulation data.[Key words] moving window extreme le