Constructive Neural Network Time Series Prediction In Gas Applications

In recent years the application of data mining is faster and faster, but when facing mass data, high dimensional data, distributed data or dynamic data, traditional methods of data mining can not acquire knowledge apace and effectively. The structure of granular computing and neural networks to deal with incomplete, ambiguous, complex, high dimensional, massive data, has its own unique features and advantages. So some aspects of the corresponding works has been done in this paper.(1)For the problem of high dimensional, massive data mining, the paper analyzed the characteristics of granular computing, summarized the application of granular computing in various fields of data mining, including application in generalized knowledge mining ,classification, clustering, association rules and other types of knowledge mining, and summed up the advantages of granular computing applied to data mining.(2) Traditional neural network to solve large-scale practical problems have great difficulties in the shortcomings of a constructive neural network edge in this regard. Analysis of the shortcomings of the commonly used covering algorithm, commonly used algorithm is based on covering all the training samples are accurate assumption, and does not take into account the data discussed the situation with imprecise. If the direct application of the method inaccurate data conditions, which are not ideal. Covering algorithm is applied to the prediction of gas proposed in this paper an improved covering algorithm, and the covering algorithm combined with the time series, applied to the gas time series prediction.(3) Gas prediction model was established,Size of the model in the theoretical framework of commercial space, with a constructive neural network learning method of gas concentration prediction. Using quotient space granular computing theory, the problem can be macro-level analysis - examining different particle size between the quotient space conversion, movement, interdependent relations, and the original features of the database information to build grain size, using a variety of granularity, from different levels of analysis of complex gas data makes the learning characteristics of the sample is more obvious, in order to better meet the requirements of machine learning. Constructive neural network learning method can be different from the micro-size structures on the quotient space data mining. At last, the method is applied to predict gas concentration, and the satisfying results are achieved. It is expected that Constructive Neural Network Learning Method will have wide applications.