Research On Node Localization Algorithm In Wireless Sensor Networks Based On Machine Learning

As an important practical application, the technology of node localization is one supporting technology of wireless sensor network (WSN), get great attention by related companies and researchers, but so far, very mature node location technologies are not recognized by the industry. As a learning method, Machine Learning, uses the experience to improve its performance. Using the characteristics of Machine Learning algorithms, to locat the nodes in wireless sensor network. The basic idea is that:the network area is divided into several equal portions of the small grids, each small grid represents a certain class of Machine Learning algorithm. After the Machine Learning algorithm has learnt the classes corresponding to the known beacon nodes, to classify the unknown nodes’ localization, and then to further determine the coordinates of the location of unknown nodes.In this paper, I mainly do the following tasks:1. For the node localization algorithm of Support Vector Machine based on ranging, generate the distance vectors between in all the beacon nodes’ distances, and use it as the training data of SVM, while adopt the distance vectors between unknown nodes and beacon nodes as testing data. After training, the SVM can classify the unknown nodes, so we can get the unknown nodes’coordinate information through regional classification.2. For the range-free localization algorithm based on SVM decision tree, in the training phase, we organize all the classes in the network area into a binary decision tree, on this basis, determine the classification, and then to determine the location information of unknown nodes. At the same time, for the case of classification wrong path in the decision tree classification process, we do the error analysis. 3. Then we do the simulation experiments verification for the location algorithms mentioned above. For the SVM OneAgainstOne Location Algorithm, the experiments show that this methord has a high location accuracy and a better tolerance for the ranging error, while it doesn’t require a high beacon node ratio, so it is suitable for the nerwork environment where the beacon nodes is sparse. For the SVM Decision Tree Location Algorithm, we do the simulation experiments in the following cases:no coverage holes in network area and existing coverage holes in network area. The results show that this algorithm is not affected seriously by coverage holes, it is suitable for the network environment of nonuniformity distribution or existing coverage holes.