Research On The Technology Of Communication Network Data Processing Based On Bayesian Theory

Probabilistic method, represented by Bayesian Theory, is now one of the hot issues in academic circle, concerning the uncertainty of network data transmission. This thesis explores and studies the problems raised by such uncertainties, mainly focusing on several typical problems appeared in the process of data transmission and compression of wireless network which have wireless sensor network technology as its the major form, seeks the solution and the model algorithm, and finally gives the theoretical and experimental verification, under the framework of Bayesian Theory.When it comes to solving Bayesian theoretical simulation, this thesis shows interests in Bayesian network, one of the hotspots in artificial intelligence algorithms, which is of deep theoretical accumulation and has been applied successfully in some fields. However, on the other hand, as a technology still in the development stage, Bayesian network itself has plenty room for refinement and improvement. For that reason, this thesis first puts forwards some suggestions for Bayesian network to make it more suitable for wireless sensor networks, and then under the framework of Bayesian Theory, proposes the error-tolerant transmission scheme and error-tolerant reconfiguration for the network data.This paper has mainly studied the following content:1. With the theory of information, the inward redundancies of the network transmission data has been analyzed comprehensively from the PME(principle of maximum entropy) perspective.It has been pointed out that the key to promote the communication quality and efficiency is to make effective use of the transmission data redundancy. From the perspective of information entropy and conditional entropy in the information theory, the relationship between uncertainty and data value heterogeneity as well as data statistics relevance has been analyzed, and it has then been pointed out that data value heterogeneity and data statistics relevance are natural redundancy in the data communication. Based on this, the core problem of his paper, how to promote the communication quality and efficiency with the natural redundancy, has been proposed.2. With the Bayesian Theory, the network data processing uncertainty has been modeled, and results of Bayesian reasoning in different probability spaces has been analyzed theoretically with certain evidences.By analyzing the research outline, general modeling of uncertainty issues of network transfer data has been discussed under the Bayesian Theory. Moreover, Bayesian network have been studied focusing on the prior probability expression, network learning and probabilistic reasoning. Based on that, with manifestation forms of Bayesian reasoning in different kinds of probability space, the condition has been proved and analyzed specifically on which the reasoning conclusions under the joint probability model and the marginal probability model are equivalent. The conclusion can provide guidance ideas for the engineering application.3. Three fast Bayesian network learning methods have been proposed to address the problem of slow convergence speed with the Bayesian network learning algorithms presently.Firstly, a new hybrid learning algorithm (SBSA) has been proposed for dealing with the static network. Started from analysis of mutual information btween the random variables, the method to construct the target network structure space boundary has been proposed and has been proved completely. According to the experiment result, the search algorithm on the base of structure boundary constraint has the better execution efficiency.Secondly, after improvement of the structure learning algorithm for the static Bayesian network, the method of transfer learning has been introduced for the problem of time-variant protogenesis distribution. Structure information and sample space information used for transfer learning by new and old models have been analyzed theoretically. From the angle of certain definition equivalence, the structure transfer learning algorithm has been proposed. Then, from the angle of probability space isomorphism conversion, the sample space transfer learning algorithm has been proposed.Lastly, effects of these two transfer algorithms have been evaluated through simulation. The structure transfer learning algorithm needs only less than 10% of the calculation amount of the static learning algorithm to finish transfer learning under the new sample, while the sample space learning algorithm has also showed the good new data adaptability during the continuous switching of samples.4. Taking into account that the node’s communication ability is limited, a error-tolerant transmission theoretical model is founded to addressed the problem with communication efficiency decreased which is caused by failibility of transmission data in networks. The method achieves maximum retention of useful information by partical error correction and particial delivery as well as transfering computational quantity to the sink nodes which has more computational abilites.,so that the efficiency is improved overall. A new coding-free forward error correction method for protocol fields has been proposed, which implements paylords aggregation effectively and provided the theories foundation for error-tolerant transmission as well.The wireless sensor network has been selected as the application target. With its feature of the simple protocol structure, feasibility of the coding-free forward error correction means has been proved through theoretical research and simulated verification by analyzing redundancy of the protocol control fields in the grouped data. On this base, the posterior forward error correction model based on the Bayesian network has been proposed. Then, for the time-variant characteristic of the prior probability model, the more practical error correction model based on the short-term statistics has been proposed in combination with the transfer learning methods in this paper, and the simulated experiment has proved that the algorithm has the higher adaptability.5. Payloads may be incorrect under the framework of error-tolerant transmission, which leads to difficultis in information recovery and extracting. Dealing with this problem, considering redundancy of the payload field, A error-tolerant compressed sensing restructuring algorithm, which is combined with the Bayesian compressed sensing and Kalman recursive least squares structure, has been proposed. The algorithm can restructure data from the incorrect ones with sufficient accuracy,. Moreover it improve the error-tolerant transmission theoretical model further.For redundancy of the transmission data business payload fields in the wireless sensor network, the error-tolerant compressed restructuring algorithm represented by the recursive least squares and the dynamic prior knowledge has been proposed based on the Bayesian compressed sensing and Kalman filtering theory. The algorithm has been verified experimentally in the wireless sensor network multi-target location application, which turns out that the algorithm has greatly promoted the error tolerance of the restructuring algorithm by introducing the prior model and retaining the high compression ratio gathering ability of the traditional compressed sensing.In the end, the paper is enclosed with a complete summary of the reseach work which has been done. Especially, the network data error-tolerant transmission theoretical model has been summarized concisely. The significance of the research is discussed as well as the shortcomings. Moreover the future research direction is indicated, from a personal perspective.