论文部分内容阅读
集成建模方法能显著提高软测量模型的预测性能,其中选择性集成通过剔除一些性能不佳的子模型,能进一步提高整体软测量模型预测性能。针对目前选择性集成研究中因忽略了数据间的差异性而导致模型预测性能不佳的问题,提出了一种动态选择性集成神经网络软测量建模方法。首先将原始数据集分为训练集和验证集,采用bootstrap算法对训练集进行差异性扰动,建立了多个神经网络子模型;然后对每个待测样本,采用K-最近邻搜索算法从验证集中找到一个最近邻子集,用该子集评估各神经网络子模型的预测性能,为待预测样本筛选合适的神经网络子模型;最后根据各子模型的预测性能合理分配组合权重,从而建立集成模型,并实现待预测样本的预测。将该建模方法应用于聚丙烯熔融指数软测量研究中,研究结果表明:与单一神经网络、常规全集成和静态选择性集成神经网络模型相比,基于动态选择性集成神经网络的熔融指数软测量模型具有更佳的预测精度。
The integrated modeling method can significantly improve the prediction performance of the soft-sensing model. The selective integration can further improve the prediction performance of the overall soft-sensing model by eliminating some sub-models with poor performance. Aiming at the problem that the performance of model prediction is not good due to the neglect of the differences among the data in the current selective integration research, a dynamic selective integrated neural network soft sensor modeling method is proposed. First, the original dataset is divided into training set and verification set. The bootstrap algorithm is used to make the perturbations of the training set. Several neural network sub-models are established. Then, the K-nearest neighbor search algorithm is used to verify each sample A subset of neighbors is concentrated to find out the prediction performance of each neural network sub-model, and the appropriate neural network sub-model is selected for the sample to be predicted. Finally, according to the predictive performance of each sub-model, the combination weight is rationally allocated to establish the integration Model, and to achieve the prediction of the sample to be predicted. The modeling method is applied to the soft measurement of polypropylene melt index. The results show that compared with the single neural network, the conventional fully integrated and static selective integrated neural network model, the melt index based on dynamic selective integrated neural network The measurement model has better prediction accuracy.