The Research Of Dynamic Data Mining Based On Neural Network

The traditional methods of Data Mining , analysising and possessing information only on historical data although it is already widely used in modern society. People no longer just found the hidden law in the historical data to solve practical problems, but get instant useful information in the highly competitive society, this traditional data mining analysising on the historical and static data is not very good to achieve such needs. Design a dynamic information processing technology aiming at current dynamic data analysis has great practical significance. Data Prediction is one of the major aims of Data Mining, the predection for the multi-dimensional is the urgent key issues need to resolve, to a certain extent, this has become a bottleneck. We research on dynamic multi-dimensional data prediction combination of Dynamic Data Mining has broad application of practical significance. Dynamic Data Mining is not applied only to predection , researching its application also has great practical significance.The author analyzes the status of the already-existed information processing technology and present a new information processing technology which combines historical data, current data and incoming data and then analyzes and processes them during the Dynamic Data Mining. The author use the Sliding Window accessing data dynamically, and process the data through the Dynamic Data Window, and use the future data testing the performance of Dynamic Data Mining. As a model that simulates human brain's work, Artificial Neural Network with a powerful parallel processing capability and memory function is now widely used in all fields. The author probes into multi-dimensional data prediction through combination of Neural Network technology,and has made substantial breakthroughs: several common Neural Network prediction models are under analyses and a single-input-and-single-step Dynamic Data Prediction Model and a single-input-and-multiple-step one and a multiple-input-and-multi-step one based on Recurrent Neural Network are brought forward. The network has a good memory because its hidden node has one step delay feedback. As its hidden node of attribute has one step inter-delayed between the association attributes, The model has very good stability and practicality. The author analyzes the Improvement Back-Propagation Algorithm and combines with the prediction model and takes full account the performance of error function, the activation function and the learn rate,And brings about his original idea on improving Back-Propagation Algorithm. Also the author testifies the possibility of prediction mode based on improving combining Back-Propagation Algorithm according to Matlab emulational test, experiments show that: the three multi-dimensional Dynamic Predection Model forecast effectivelly. As multi-dimensional curves prediction,The effect is particularly good. As multi-dimensional line prediction,the prediction-error is zero.the model of good stability is not impact on the initial weight. In this paper, a framework concept to achieve update knowledge real-time of the Expert System is proposed.The Dynamic Prediction Model,proposed in this paper,considers the internal relations on each attributes of the multi-dimensional data,the multi-dimensional data dynamic prediction is more appropriate to the actual realization.also,to a certain extent,it provides a method for further researching on dynamic data of multi-dimensional.