Study On Model And Algorithm Of Dynamic Feature Fusion Based On Information Sources Selection And Sequential Extraction

In this paper, we try to study information fusion model on character level, and then give an algorithm and application example to support the model. First we introduce a descriptive definition of data∕information fusion , application of this technology, typical fusion methods and research field internal and international. We also developed the concept about the fusion level. Then we discuss the architecture and technology about character fusion, assume a new idea and project of technical problem——how to select valuable information from dynamic information that increasing and changing rapidly for fusion . Facing huge dynamic information "heaps", a complete fusion mechanism should be able to select valuable information from all. The mutuality between information sources and object, must be considered, also the time-effective of information. Only valuable data joint to fusion. Being inspired by the information selecting structure of KDD/DM and thinking of co-operation and co-complement of various spatio-temporal information , we find a new mechanism which is effective in KDD∕DM and try to replant to fusion model, so that the new mechanism can find essential information sources and time period from continuous information providing by multi-source without being told which data is useful or not. Furthermore, we give a idiographic model by using this kind of mechanism which can gathering dynamic data by using the slice sequence, picking-up strategy of self-selection of information sources and slice series and reconstructing dynamic examples, propose a spatio-temporal fusion model based on "two-space model"which adjusts itself to gain complete experience knowledge. We also emphasize concurrency, utility and realization thinking of both Discovery∕Mining and co-operation∕co-complementation. Finally, we discuss and analyze uncertainties in detail, advance a solution strategy of improving compactness and decreasing uncertainties in characteristic level fusion . Then, a new algorithm is designed to descript selection fusion system. When learning from examples, this algorithm regards information gain as scale of the fusion mechanism selection. We also find out that scaling the stability of fusion system by information gain is better than by entropy. This algorithm improves Quilin's ID3 (a decision-tree algorithm) in order to fit dynamic association. It can fuse multi-series, select relation rules and wash out false elements. The new algorithm has ability of processing dynamic multi-parameter. Besides of character value, it can fuse procedure and dynamic correlation then give comprehensive estimation. In order to test the effectiveness of the algorithm, we have applied it to the hydropower station. After analyzing the particularity of the generate electricity with waterpower, we apply our model to supervise the running general transformer, and the simulating test supports this model and this algorithm. Comparing traditional fusion model, our achievement is that the new model can tell us if the collection information is useful or not. So it is significant in the information "exploding"age, and can select relative sources and period in many noisy sources environment and continuously series. Experiment test has got a significant result.