多阶段逆模型方法及在胶液生成过程中的应用

Multistage Inverse Modeling Method and Its Application in Gelatin Solution Production Process

针对一类串联型工业大系统,提出了多阶段逆模型建模方法:将串联大系统分为若干个阶段,以产品质量指标作为过程设计的起点,用逆向推理的方法,建立各个阶段的逆模型;根据产品质量指标的要求,直接求出各个阶段的控制变量设定值。将该方法应用于胶液生成过程的软测量建模,采用多阶段建模方法和整体建模方法分别建立了基于BP神经网络的胶液生成过程逆模型,并从误差平方和MSE和命中率等方面对两种建模方法的建模精度进行了比较。结果表明,多阶段建模方法可以获得更高的建模精度;同时,具有更大的灵活性;而且逆模型方法可以根据质量指标求出控制变量设定值,更便于实际应用。

Aiming at a class of serially connected industrial system, a novel multistage inverse modeling method was presented. The large-scale system is divided into several stages. Using specified product qualities as a starting point for process design. By backward reasoning the required process conditions and the control variable set points of all stages for processing system were found. The inverse models of gelatin solution production process were established based on the BP neural network by using multistage modeling method and whole stage modeling method, and modeling accuracy comparison were made from error and hit rates. The simulation results indicate the model based on the proposed method has smaller error and higher hit rates. Meanwhile, the break down of the sub models increases the flexibility of model development and reduces the effort to change the model when the sub models change. And the required process conditions and the control variable set points of all stages for processing system were found according to specified product qualities. Thus, it is easy to be really applied. This method has been successfully applied on improving the gelatin solution production process and product quality control.