中文摘要：

信息-物理融合系统（CPS）由物理事件与计算系统两大部分组成,前者专注于处理连续的物理过程,与时间的流逝存在密切联系;后者则只能处理基于0-1机制的离散问题,两者存在本质上的差别.传统的建模方法一般需事先估计系统将要经历的状态及每一个状态转换所需的时间,这在复杂的CPS（Cyber-Physical System）环境中难以预测,也没有真正意义上实现计算系统与物理事件之间的交互.因此,我们提出一种新的建模方法 DCDM：面向CPS的真实环境经传感器、网络等测试得出一些原始的离散数据,通过一定的数学方法（GMDH）从数据出发建立系统的连续模型,为了缩小离散数据与连续模型之间的误差,采用反馈控制的方式不断调整两者之间的差距,直至其减小到一定的范围之内.DCDM从根本上改变了传统的建模方式,提出了一种从离散数据出发建立系统模型的思想,通过反馈控制实现物理事件与计算系统之间的深度融合.DCDM具有以下4种优势：（1）建模对象容易获得且更加客观,能真实反映系统本身;（2）通过数学方法直接实现参数的筛选,去除冗余属性,减小弱影响因子对系统模型的影响;（3）相较于其他数据拟合的方法,DCDM具有更加简单的复杂度,能节约大量的计算时间;（4）在离散数据与连续模型之间的误差方面,相较于其他的算法有着明显的改进.实验结果表明,在真实数据集中,DCDM的执行速度快于当前最新的数据建模方法,且数据集越大优势越明显.

英文摘要：

Physical process and computing system are two aspects of CPS.The former focuses on processing continuous objects which requires time-dependent models,and the latter takes its attention in handling the discrete problems （0-1 formalism）in most cases,so it exists great difference in modeling method of this two aspects.Notice that the traditional modeling methods focused on states transition in general,which is difficult to predict in such a complicated environment of CPS,it didn’t truly realize the interaction between the two aspects of CPS.In this paper,we obtained some discrete data from the real environment,and presented a new modeling mechanism called DCDM which started from the original data to build a continuous model through a mathematical method （GMDH）.To decrease the relative error between them,we took the feedback control to the original data to adjust the error.Furthermore,from the adjustment of new discrete data,our method established the continuous modeling again by repeating the above steps until the error turns to an acceptable range.DCDM has fundamentally changed the traditional modeling style with the proposal of a new modeling direction which models the system from discrete data,and deeply realizes the integration between the discrete system and continuous system through feedback. DCDM has the following four advantages：（1 ）the modeling object is easy to be obtained and objective to reflect the system itself;（2 ）DCDM can directly select the suitable parameters through the GMDH to remove the redundancy attributes and minimize the factors impact on the system models;（3）Compared to other data fitting methods,DCDM has a simpler complexity to save a lot of computing time;（4）There is a significant improvement in the error between the discrete data and the continuous model.Experimental results show that the implementation of DCDM is faster than the current data modeling method,and it is more obvious with the increase of data.