论文部分内容阅读
工业过程的模型通常具有非线性强、系统时变明显、工况变化大等特点。传统的滑动时间窗选择方法容易包含大量关联度低的数据,影响了数据模型的建模精度和建模效率。本文提出了一种带状态约束的滑动时间窗口选择算法,应用于工业连续过程软测量模型的训练数据选取。将该时间窗口选择算法与最小二乘支持向量回归算法结合,利用电站锅炉的历史运行数据,建立了燃煤电站锅炉尾部烟气含氧量的软测量模型。研究结果表明,相比于传统的滑动时间窗选择算法,利用该算法进行训练样本选取后,提高了所建立的最小二乘支持向量机模型的模型精度和运行效率。利用该方法建立的烟气含氧量软测量模型具有较高的精度,可以在氧化锆传感器发生故障时代替其工作,保证了氧量信号的稳定性和可靠性。
The model of industrial process usually has the characteristics of strong nonlinearity, obvious time-varying system and large change of working conditions. The traditional sliding time window selection method easily contains a large amount of data with low relevance, which affects the modeling accuracy and modeling efficiency of the data model. In this paper, a sliding time window selection algorithm with state constraints is proposed, which is applied to training data selection of industrial continuous process soft sensor model. Combining the time window selection algorithm and the least square support vector regression algorithm, the soft sensor model of the oxygen content in the tail flue gas of a coal-fired power station is established by using the historical operating data of the boiler. The results show that compared with the traditional sliding time window selection algorithm, using the algorithm to select the training samples, the model precision and operating efficiency of the least square support vector machine model are improved. The soft sensing model for flue gas oxygen content established by this method has high precision and can replace the work of zirconia sensor in the event of a fault, ensuring the stability and reliability of the oxygen signal.