改良ANN-BP算法在炭黑工艺建模中的应用与研究

Application and Research of Meliorated ANN-BP Arithmetic in Modeling for Carbon-Black Technics

针对目前炭黑行业生产主要以经验为主的不利境况,将人工神经网络误差反向传播算法(ANN BP算法)用于炭黑工艺建模,比较三种ANN BP算法结果后,利用基于动量法和学习速率自适应调整改良的ANN BP算法建立了炭黑工艺参数与指标之间的非线性映射模型,并与多元线性回归、主成分回归建立的线性模型进行了比较.结果表明,改良ANN BP算法预测相对误差在5.6%以内,且有较好的容错能力,比较好的解决了炭黑生产过程中的预测模型构建问题.

In allusion to the disadvantage condition on the production of carbon black by means of experience principally, and the scarcity of perfect prediction model, meliorated artificial neural network with error back-propagation (ANN-BP) is applied to the modeling of carbon-black technics for the first time. After the comparison for the results from three types of ANN-BP,meliorated ANN-BP,based on momentum method and learning rate by self-adaptive modulation, is applied to set up the nonlinear mapping model between parameters and targets, consequently, the mapping model is compared with the linear models based on multiple linear regression and principal component regression. The results indicate that the relative prediction error based on meliorated ANN-BP arithmetic is under 5.6%, the model with preferable accommodation error is suited to solve the problem of setting up prediction model of carbon-black production.