基于混合高斯回归的动态软测量方法研究

发布时间：2018-09-13 06:43

【摘要】：工业过程中,许多关键变量难以在线测量。软测量技术成为解决该问题的关键技术之一。目前,许多模型是基于测量过程处于稳态工况的假设,故模型大多为静态软测量模型。由于生产过程中工艺改变、物理结构特性变化以及环境等因素的影响使得生产过程通常处于动态工况,仅考虑对象静态特性是不够的,有必要对动态软测量建模方法进行深入研究。本文以青霉素仿真过程和工业红霉素发酵过程为对象,采用高斯过程回归以及混合高斯回归方法来研究建立动态软测量模型。具体工作如下:(1)针对静态软测量模型预测精度低和鲁棒性差等问题,本文提出了一种基于多准则和高斯过程回归的动态软测量建模方法。该方法综合考虑多种模型定阶准则,提出了高斯过程回归动态软测量模型定阶策略,为模型阶数确定提供了依据,并将所提动态软测量模型应用于红霉素发酵过程的生物量浓度估计。研究结果表明,基于高斯过程回归的动态软测量建模方法可以实现对生物量浓度的高精度预测,且预测结果具有较小的置信度区间。(2)针对单高斯过程回归动态软测量模型的局限性,提出了一种混合高斯回归动态软测量模型。该模型有两个重要参数,即高斯元的个数和模型阶数,为了获得优化软测量模型,提出一个迭代策略优化两个结构参数。将所提出的动态混合高斯回归软测量模型应用于青霉素仿真过程和工业红霉素发酵过程的生物量浓度估计,并和现有的动态高斯过程回归软测量模型比较。结果表明,所提动态混合高斯回归软测量模型具有较高的预测精度,更适合于动态多相/多模态发酵过程。(3)针对移动窗口算法和混合高斯回归软测量模型中移动窗口算法的大量更新,降低了模型的计算效率,并占用大量的计算机内存资源等问题。本文提出了一种基于模型性能评估的递推混合高斯回归建模方法来减少递推混合高斯回归建模方法的模型校正频率。首先,根据过程初始特性自动生成模型的初始置信限。然后,将预测均方根误差指标作为评价模型的标准。根据模型的性能评估结果,选择性地激活模型校正,同时更新置信限。最后,将开发的模型用于青霉素仿真过程和工业红霉素发酵过程的生物量浓度软测量。仿真结果表明,所开发的模型大大提高了计算效率(模型校正频率大大降低),而预测精度的损失可以忽略不计,且与混合高斯回归模型相比,预测精度明显提高。
[Abstract]:In the industrial process, many key variables are difficult to measure online. Soft sensing technology has become one of the key technologies to solve this problem. At present, many models are based on the assumption that the measurement process is in steady state, so most of the models are static soft sensor models. Due to the influence of process change, physical structure characteristic change and environment factors, the production process is usually in dynamic condition, so it is not enough to consider the static characteristics of the object. It is necessary to study the modeling method of dynamic soft sensor. In this paper, the simulation process of penicillin and the fermentation process of industrial erythromycin were used to establish dynamic soft sensor model by using Gao Si process regression and mixed Gao Si regression method. The main works are as follows: (1) aiming at the problems of low prediction accuracy and poor robustness of the static soft-sensor model, a dynamic soft-sensor modeling method based on multi-criteria and Gao Si process regression is proposed in this paper. This method synthetically considers several model order determination criteria, and puts forward the order determination strategy of Gao Si process regression dynamic soft sensor model, which provides the basis for determining the model order. The dynamic soft sensor model was applied to estimate the biomass concentration of erythromycin fermentation process. The results show that the dynamic soft sensor modeling method based on Gao Si process regression can achieve high precision prediction of biomass concentration. The prediction results have a small confidence range. (2) in view of the limitation of single Gao Si regression dynamic soft sensor model, a hybrid Gao Si regression dynamic soft sensor model is proposed. The model has two important parameters, namely, the number of Gao Si elements and the order of the model. In order to obtain the optimized soft sensor model, an iterative strategy is proposed to optimize the two structural parameters. The dynamic mixed Gao Si regression soft sensor model was applied to estimate the biomass concentration of penicillin simulation process and industrial erythromycin fermentation process. The results show that the proposed dynamic mixed Gao Si regression soft sensor model has high prediction accuracy. It is more suitable for dynamic multiphase / multimodal fermentation process. (3) aiming at a large number of updates of moving window algorithm and mixed Gao Si regression soft sensor model, the computational efficiency of the model is reduced. And occupy a large number of computer memory resources and other problems. In this paper, a recursive mixed Gao Si regression modeling method based on model performance evaluation is proposed to reduce the frequency of model correction of recursive mixed Gao Si regression modeling method. First, the initial confidence limits of the model are automatically generated according to the initial characteristics of the process. Then, the root-mean-square error index is taken as the criterion of the evaluation model. Based on the performance evaluation of the model, the model correction is selectively activated and the confidence limit is updated. Finally, the developed model is applied to the soft measurement of biomass concentration in penicillin simulation process and industrial erythromycin fermentation process. The simulation results show that the developed model greatly improves the calculation efficiency (the calibration frequency of the model is greatly reduced), and the loss of prediction accuracy can be negligible, and compared with the mixed Gao Si regression model, the prediction accuracy is obviously improved.
【学位授予单位】：江苏大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：O212.1

【参考文献】