烟气含氧量软测量新方法研究

The Study of a New Method Incorporating the Soft Sensing of Oxygen content in Flue Gases

针对目前大量工业现场使用的氧量分析仪成本昂贵、维护费用高且易损坏等问题,在几种常见方法对比讨论的基础上,提出了利用基于统计分析和神经网络技术的NNPLS方法建立烟气含氧量软测量模型的方法。该方法综合了PLS和神经网络技术的优点,能够利用过程历史数据辨识对象模型;利用现场实际数据对该方法进行了仿真验证,并将仿真结果与传统的线性PLSR方法和直接神经网络建模方法作了比较,结果显示NNPLS方法所建立的软测量模型具有更强的泛化能力。文中还对静态模型向动态模型进行了扩展。

In view of the high first cost of conventional oxygen-content analyzers for industrial applications, their high maintenance expenses and low durability the authors have on the basis of comparing several commonly used methods come up with a new method for measuring oxygen content in flue gases. The proposed method involves an oxygen-content soft sensing model set up through the use of a NNPLS (neutral network partial least square) approach based on statistical analyses and neural network technology. It enjoys both the merits of PLSR (partial least square regression) and neural network technology, making it possible to identify a target model by utilizing historical process data. A simulation verification of the method has been conducted by using on-site industrial data. In addition, the simulation results are compared with traditional linear PLSR method and the direct neural network-based modeling method. The resultS of comparison indicate that the soft sensing model based on the NNPLS approach features a more effective generalizing ability. Furthermore, an extension of a static model to a dynamic one was also performed.