欢迎您！东篱公司
退出
-
申报数据库
-
立项数据库
-
成果数据库
-
项目获奖数据库
中文摘要：
针对具有多工况特征的工业生产过程,多模型建模是一种有效的软测量建模方法。建模过程中,聚类方法、建模方法及融合方式都会对模型的精度产生影响。因此,提出一种改进自适应仿射传播聚类的多模型建模方法。首先,采用自适应仿射传播聚类算法确定偏置参数近似值,并用差分进化算法对偏置参数和阻尼系数进行局部范围内寻优,划分得到更优的子数据集;然后,建立各个高斯过程回归子模型;最后,对于新来的数据,利用贝叶斯融合方法自适应地计算出各子模型的权重,融合各子模型预测值得到最终的输出。通过对标准数据集和青霉素发酵过程数据的建模仿真,验证了所提方法的有效性。
英文摘要：
Multi-model soft sensor modeling methods are effective for multi-mode industrial processes. During the modeling process, the prediction accuracy can be influenced by clustering strategies, modeling approaches and ensemble learning methods. A multi-model soft sensor method is therefore proposed in this paper based on improved adaptive affinity propagation clustering algorithm. Firstly, an adaptive affinity propagation clustering algorithm is applied to determine the approximate value of preference parameters, and the differential evolution algorithm is then employed to optimize the preference and damping parameter on a local scale, thus more accurate sub-datasets can be obtained; Afterwards, the Gaussian process regression method is utilized to construct sub-models; Finally, for the new data, the Bayesian ensemble method is adopted to calculate the weights of sub-models adaptively, the final prediction result is obtained by combination of different models＇ performance. The simulation results of a standard dataset and soft sensor modeling of the penicillin fermentation process altogether indicate the effectiveness of the proposed algorithm.
同期刊论文项目
同项目期刊论文
Effectof temperature on Chinese rice wine brewing with high concentration pre-steamedwhole sticky ri
期刊信息
位置：