Research And Applications On The Key Technology Of Multi-source Heterogeneous Data Fusion

People's requirement for abundance,real time performance,accuracy and reliability of information can't be satisfied with single-source data,there comes the data fusion technology who can draw an estimation and judgment from multi-source data,which could improve the reliability and reduce the uncertainty of data.This thesis introduces basic situation and key technology of data fusion firstly and take Dempster-Shafer Evidence Theory as main research object,aiming at the difficulty of obtaining Basic Probability Assignment function of evidence theory,a fuzzy weighted data fusion framework based on D-S Evidence Theory and a corresponding data preprocessing algorithm on denoising are proposed.The main work and innovation of this thesis are listed as follows:1.A data-denoising method based on FCM algorithm is proposed.Data preprocessing is an essential stage of data fusion,in this thesis,based on the problem of data preprocessing noise data,starting from the measurement of data similarity,using Cosine Similarity to construct weighted values of Euclidean Distance,a data-denoising algorithm based on FCM is constructed.Experimental results show that this algorithm is superior to some classical algorithms in noise detection rate.2.A fuzzy weighted data fusion framework based on D-S Evidence Theory is proposed.For expert system's lack of objectivity and typical sample method's difficulty on obtain valid confidence intervals and some other issues,a compound model is constructed,which take fuzzy Naive Bayesian as generative method and FCM as discriminative method.For the uncertain information,the model use a reliability allocation method to determine its mass function and find the weights of generative BPA and discriminative BPA through a credible mathematical structure,which could attributes better effect in practical application.Then the algorithm is proved to be effective on data fusion,especially for the uncertain data.3.Design and implement a network attack identification prototype system.The improved algorithm proposed in this paper is applied to the actual attack recognition detection case,and build real network environment for testing.The system starts from data collection module,grabs data packets through the sniffing tool,and obtains the judgment of network behavior through data preprocessing and data fusion and displayed by visualization module.Experimental results show that the fusion results of the proposed algorithms are not only better than those who are generated by single source data,but also better than some classical algorithms.When there exists more uncertain information,the algorithm proposed in this thesis has better performance in false-alarm-rate and missing-alarm-rate.The attack identification system implemented in this paper can effectively identify the attack type and display the results intuitively.