贝叶斯框架下的LSSVM和贝叶斯网络及其应用

LS-SVM algorithm of Bayesian framework and Bayesian regularization network and their applications

贝叶斯理论能够利用样本信息和先验知识,简化预测模型,优化参数。主要介绍了贝叶斯框架下的最小二乘支持向量机算法和贝叶斯正则化神经网络,贝叶斯框架下的最小二乘支持向量机能确定正则化参数和核参数,贝叶斯正则化网络能够自适应的调整网络的复杂度和网络的隐节点个数。以轻柴油的凝点、闪点、95%馏出温度3个关键指标输出为例分别建立了这两种预测模型,并且对结果进行了比较,仿真结果表明贝叶斯框架下的最小二乘支持向量机比贝叶斯正则化网络有更强的泛化能力,而且程序运行速度快,运算精度高。

Bays theory can make use of sample information and prior knowledge, simplified prediction model and optimized parameters. Lease support vector machines algorithm of Bayes frame （Bays LS-SVM） and Bayes regularization neural network （Bays NN） is introduced. Bays LS-SVM can determine the regularization parameters and kernel parameters effectively, Bays NN can adjust the complexity of network and the hidden nodes adaptively. Two kinds of prediction model of light diesel oil are built. The simulation result indicates Bays LS-SVM has stronger generalization ability, faster procedure speed, and higher operational precision.