基于支持向量机的干气制乙苯反应器出口温度预测模型研究

The study on temperature prediction model for dry gas-to-ethylbenzene reactor's outlet based on Support vector machine

鉴于化工过程往往机理复杂、耦合性强、高度非线性,难于建立其机理模型,这时就需要采用经验建模的方法。支持向量机是一种新的机器学习方法,其基于结构风险最小化原则,用支持向量机建模不需要考虑对象机理,且对非线性问题有很好的效果,是一种良好的经验模型,己被应用于不少化工问题中。在本文中我们把支持向量机这一新颖算法应用于干气制乙苯反应器出口温度预测模型中,简要介绍了支持向量机的一些基本理论,在此基础上详细研究支持向量机在干气制乙苯反应器出口温度预测模型建模中的应用。首先,选择支持向量机的类型为ε-SVR,通过四种核函数在实际预测中误差的比较选择径向基(RBF)核函数作为本文支持向量机模型所用的核函数,之后应用交叉验证的方法选择最佳参数C=4,γ=0.0051543,最后建立预测模型并对训练集和预测集分别预测,预测结果相关系数在90%以上,说明模型精度达到要求。对支持向量机和遗传算法优化的BP神经网络算法的建模效果进行综合比较和讨论,得出支持向量机与传统建模方法相比有更好的预测准确率的结论。

Since the mechanism of chemical processes are often complex,strongly coupled,highly nonlinear and difficult to establish the mechanism model,a method using empirical modeling is required.Support Vector Machine is a new machine learning method,which is based on the principle of structural risk minimization,support vector machine modeling is a good experience model as it does not need to consider the mechanism of the object,and it has a good effect on the solution to nonlinear problems.It has been applied to many chemical problems.In this paper we use support vector machine algorithm to the temperature prediction model of dry gas-to-ethylbenzene reactor＇s outlet,and the basic SVM theory is introduced briefly,support vector machines in the dry gas-to-ethylbenzene reactor＇s outlet temperature prediction model modeling was studied in detail on this basis.At first,the type of support vector machines wasε-SVR.Through comparison of the error of four kernel functions in the actual prediction,radial basis（RBF） kernel function was selected use for support vector machine model in this paper, then cross-validation method is used to select the best parameters C = 4,γ= 0.0051543,at last,establish the prediction model and predict the training set and prediction set respectively,the correlation coefficient of predicted results are above 90%,indicating that the model＇s accuracy can meet the requirements.Finally,support vector machines and GA-BP algorithm for modeling are compared and discussed comprehensively, come to a conclusion that support vector machines has better prediction accuracy in comparison with traditional modeling method.