基于贝叶斯-神经网络的动态回归建模与预测

结合贝叶斯网络和神经网络，提出了一种建立数据驱动型的动态线性回归系统模型的方法。基于这种模型采用自然连接型的知识分布，形式化各种各样的信息，结合贝叶斯方法，执行贝叶斯网络的持续学习过程；采用指数寿命型的连接权值改进径向基神经网络，优化输入数据，提高计算速度；采用改进的遗传算法，实现神经网络的动态自适应。基于上述方法，实现了线性回归系统动态建模与实时预测。仿真试验说明该方法是有效性。

Dynamic Linear Regression Modeling and Real-time Prediction Based on Bayesian Network and Neural Network

A new method was represented to model dynamic linear regression system driven by data, in which a bayesian network was combined with the RBF neural network. Based on this model, the knowledge distribution of nature connection tied in bayesian method was used to formalize all kinds of information and implement the durative process of learning. An improved RBF neural network with the exponential-longevity linked weights was used to optimize importing data and enhance calculating velocity. An improved genetic algorithm was applied to realize the dynamic adaptation of the neural network. Based on all the algorithms, a dynamic linear regression modeling and real-time predictive control system was implemented. A simulation experiment demonstrates efficiency of the method.