Research Of Localization Algorithm In Wireless Sensor Networks Based On Support Vector Machine

With the rapid development of electronic technologies and wireless communication technologies,wireless sensor networks have been increasingly widely used in many industrial areas.As the key technology of wireless sensor networks,positioning technology has achieved fruitful academic results in recent years.The traditional wireless sensor positioning algorithms often have many defects,such as high dependence on anchor nodes and poor performance under non-line-of-sight conditions,and the practicality of the algorithm is not strong.In order to solve these problems,this paper introduces the machine learning method to locate the nodes,and uses the characteristics of machine learning algorithm's generalization ability to establish node localization algorithms based on SVM and SVR,which has achieved good results.This paper studies the development status of wireless sensor positioning algorithms at home and abroad,analyzes and summarizes the basic concepts and general principles of positioning algorithms,and extracts the performance evaluation criteria of the algorithm.Through the study of the classic positioning algorithms,the advantages and disadvantages of these algorithms and the application occasions are compared.This paper introduces the positioning process of DV-Hop algorithm in detail,analyzes and studies the defects and shortcomings of DV-Hop algorithm,including using average hop distance to estimate the distance between nodes will bring large errors and the minimum hop number can not accurately reflect the relative position between nodes,and makes simulation experiments on DV-Hop algorithm.In order to solve the problem of insufficient hop estimation due to non-line-of-sight and nonlinear propagation in DV-Hop algorithm,the hopping method and machine learning algorithm are combined in DV-Hop algorithm,and MSVM dan MSVR algorithm based on hop count are proposed.MSVM transforms the location problem into a multiclassification problem by dividing the squares,and introduces a decision tree method to implement the multi-class SVM algorithm.The MSVR problem directly regards the positioning problem as a regression problem.By training the support vector regressions on two coordinate dimensions,the coordinates of the unknown nodes are directly predicted.Aiming at the problem that the minimum hops between nodes in DV-Hop algorithm can not accurately reflect the relative position between nodes,this paper proposes an HCR hop optimization method.The optimized hop count of the HCR method is a real number,and its integer part holds the original minimum hop count information,and the fractional part reflects the relative position of the node within a single hop.The HCR-SVR algorithm is constructed by using the optimized hop matrix as the input vector of the SVR algorithm.In order to further fully utilize the training data and draw on the idea of integrated learning,the EN-SVR algorithm is proposed.By using the bagging and random subspace methods to divide the training data set into multiple subsets,the SVR algorithm is used to train multiple regressionrs,and the integrated learning model is constructed.The results of these regressions are weighted and summed as the final unknown nodes.Coordinate predicted value.Besides,this paper uses Python software package to simulate the positioning algorithm,and comprehensively tests the performance of the positioning algorithm in different scenarios,and verifies the feasibility of the positioning algorithm,which has strong theoretical value and practical significance.