Automated Sample Data Selecting from DAS Based on Maximum Entropy Theory

How to selecting sample data set from a DAS automatically is a critical problem for machine learning. In this paper, it is illustrated that measurement data included enough information for modeling through comparing the information extropy of a continuous system with its sampling system. Based on maximum entropy principle, an equipartitional method has been discussed which can be used to collect a small data set as a training sample set from a DAS. Then, an application of this method which be used in an ethylene oxide reactor´s modeling has been given. The sample set obtained by this way has a uniform distribution as good as distributing of boundary data points. This application illustrates that this way was effectively for selecting a sample set. And combined with a RBF-BP cascaded artificial neural network, it got a satisfactory prediction result.