Bp Neural Network In The Numerical Simulation-based Scientific Data Mining

It is very difficult to understand or analyze the large-scale data in many scientific domains, because it has the characteristic of large-scale quantities, complex features when dealing with, and it is more difficult to get knowledge from it. So it is absolutely necessarily to do science data mining (SDM) now. The main point of this project is to research the theories and applications of artificial neural network (ANN) which is suitable for large scale science data mining. Especially, our research focus include: dimension reduction techniques based on Independent Component Analysis (ICA) and wavelet-based denoising or compressing techniques for feature extraction in scientific datasets which have complex features; Classify and clustering techniques of ANN combination with data grid , Back-Propagation neural network , Self-Growing Multilevel Self-Organizing Map for large scale knowledge founding in SDM. We propose the ANN model for special application--Molecular Dynamics numerical value simulation, especially aim at classifying, clustering and pattern cognition for scientific dataset. Construct the utility SDM system and give a new method to mine valuable information from large-scale numerical value simulation data.This paper focuses on the classification based on Back-Propagation neural network. Firstly, presents a review on the basic model, introduces the different algorithms used by the traditional BP network, then, emulation experiments have been made to compare with each algorithm by Matlab. Lastly, proposes a neural network model based on LM and evolution algorithm , which can avoid convergence to the local minimum , more quickly, and has high classification precision. The method can promote the research of large-scale pattern recognition greatly.