Neural Network Prediction And Scientific Data Mining Applications

Prediction can make a judgment for the development of the future according to the past and present situations. Based on the processing of historical data, it is important for science research. The computer technology coming is a powerful tool for collecting and processing historical data rapidly. But how to deal with thousands of megabytes of data efficiently is a great problem for all people. At the beginning of 90's, people began to extract information from seas of data using data mining technology and then it has got wider application. The task of data mining is finding patterns from data. The pattern can be divided into two categories: predictive and descriptive pattern. The predictive pattern can get certain result pattern precisely from data and the descriptive pattern is the description about a rule existing in datasets or grouping data by their similarities. So the function of data mining can be classified into two classes that are description present and prediction for future. Exact description present is the foundation of the precise prediction for future even we can say that the function of data mining is just predicting future. The data mining research has coming into a hot issue now and involved various subjects mainly including the principle of statistics, database management, artificial intelligence, machine learning, pattern recognition, data visualization and so on. Scientific data with complex feature makes it very difficult to understand, analyze and extract knowledge from them. So we need more powerful scientific data analyzing tool: Scientific Data Mining.This thesis focuses on the application of neural network prediction theory in scientific data mining from three aspects: theory, algorithm and application. With the development of high performance computer and implementation of parallel computing method in science computing fields, scientific simulation program produces seas of numerical computing data which can break through the scale of GB easily, amounting to TB or even PB level. So how to store these data effectively is a problem which needs to be solved soon. We use professional scientific data management software HDF5 to manage output data of a three-dimensional electromagnetic program for laser-plasma interaction and call compression library zlib provided by HDF5 to select suitable compression algorithm. How to select the most efficient algorithm among...