Research And Application Of Data Fusion Methods Based On Deep Learning

As the basis of network security situation awareness,multi-source data fusion technology is extremely important.In order to achieve the goals of data dimensionality and redundancy reduction,collect collaborative information and analysis comprehensively,data fusion analyzes and processes heterogeneous data from multiple sources in a comprehensive way.Due to the variety of data sources and the different structure of the data,it is very difficult to seamlessly fuse data,which makes data fusion face many problems.Therefore,only by finding an effective fusion method can we handle the intricate relationships in massive multi-source heterogeneous data and obtain more accurate fusion results.Based on carefully and comprehensively sorting out the domestic and foreign development processes and research status in the field of data fusion,this paper analyzes and compares a variety of key technologies in this field.This paper forms a new data fusion solution based on D-S Evidence Theory and deep learning algorithms.The main work and innovations of this article are as follows:(1)An improved D-S Evidence Theory method is proposed.In the case of severe conflict of evidence and the apparently unreasonable synthesis result in the case of complete conflict,and the probabilistic assignment is more subjective and unreasonable.We propose two improvements.First,based on the average support algorithm,fuzzy membership fA and support SA are introduced as weights to improve the accuracy of the composition formula.Secondly,a more accurate basic probability assignment function(BPA)is obtained by introducing a neural network with strict internal logic,good at approximating complex nonlinear relationships,and fast convergence speed.(2)When the basic probability assignment function(BPA)is obtained by training with the neural network method,huge calculation amount and calculation time will affect the efficiency,and the optimization algorithm also determines whether the neural network can converge to the optimal solution.So,we propose an improved deep learning.optimization algorithm.This algorithm introduces the concept of third-order moments into the optimization algorithm to solve the problem of unstable and extreme learning rates that may not converge to the optimal solution pair,and can accelerate the convergence process to a certain extent.(3)Based on the improved D-S evidence theory and deep learning algorithm,a scheme for detecting network attacks is designed and completed.Through the analysis of logs and traffic data of IDS,firewalls,routers,etc.,it can detect DoS attacks on network traffic and provide strong support for network security decisions.Experiments show that the improved D-S Evidence Theory algorithm,combined with deep learning optimization algorithms,is superior to some traditional algorithms,and has higher accuracy and stability;the scheme for detecting network attacks in this paper can complete the functions in demand analysis well accurately and can efficiently complete the fusion to detect network attacks.