Study On Theory And Algorithm Of Mobile Target Localization In WSN Based On LSSVR Regression Modeling

Target localization is one of the important applications of Wireless Sensor Networks and achieving reliable and accurate target localization has an important application value for the national defense and military, environmental monitoring, intelligent transportation, security surveillance and so on. Single target localization is important basis of multiple targets localization. Aiming at achieving an integrated improvements of single target localization accuracy, rapidity and reliability in physical environment with low energy consumption, the research includes local regression modeling, target prediction, nodes wake-up, localization theory of rapid modeling and localization based on Least Square Support Vector Regression (LSSVR),which has an important academic value and practical significance for the promotion of development and application of the manufacturing information technology and networked measurement control technology and strengthening cross of manufacturing engineering, instrumentation, and information subjects. The research work is funded by the Ministry of Education Support Program for New Century Excellent Talent (No.NCET -08-0211) and Guangdong Province Natural Science Foundation project (No.9151052101000013).Beginning with the basic steps of target localization in WSN, the influence relationship between various localization steps and localization performance indexes is analyzed in the paper. The research progress at home and abroad of WSN target localization, target prediction, nodes wake-up and energy consumption are reviewed generally to decide the research goals of this paper. Main research works in this paper are as follows:(1) LSSVR mathematical model and its solution method, characteristics are discussed and it is pointed out that LSSVR is suitable for complex multivariate nonlinear system modeling problem. It has clear advantage to applying LSSVR in resource-constrained embedded computing system. The basic theory and methods of LSSVR regression modeling in WSN target localization is researched and points out that there is a non-linear mapping between distance-vector and target coordinates, meeting the mathematical conditions of adopting LSSVR regression modeling. Feature vector detection nodes in terms of location-number conditions, vector space mapping conditions are researched. The target localization method based on LSSVR regression model is proposed creatively, the method has a good generalization performance in small sample, making use of LSSVR anti-noise ability can reduce the impact of localization resulted from the measurement noise. LSSVR modeling localization errors structure model is established, the model affected error and noise error reflect LSSVR regression model promotion performance and anti-noise ability respectively. Target localization error spatial distribution characteristics based on LSSVR is researched, using a reasonable regression modeling strategy can adjust the distribution function of model localization error and noise error to improve the overall results of the localization. The mechanism of effects of the kernel function for target localization based on LSSVR is discussed, the kernel function for LSSVR modeling should has a good effect with model prediction, simple form, less parameters and so on.(2)Based on the target's transmit power feature extraction conditions is analyzed theoretically. Furthermore, signal strength difference feature extraction method is proposed. The feature vectors constructed by the method has nothing to do with channel parameters P ( d 0), meeting the conditions of LSSVR modeling and localization when the target's transmit power is unstable all the same. It is able to reduce the effect of localization resulted from the target's transmit power changing. Local learning modeling idea is introduced, the distribution of the training sample points and the sampling points, the modeling region and some other rules based on the LSSVR local modeling are researched, making the method of LSSVR local modeling and localization more practically. Changes of modeling parameters effect on LSSVR local modeling and localization is researched. Moreover, the relationship between the different modeling parameters and LSSVR localization error is pointed out and compromise the parameters value is essential. Using particle swarm optimization to optimize LSSVR model parameters was proposed, the localization accuracy of LSSVR local modeling and localization is improved obviously.(3) The mathematical expression of the number of measuring nodes N d and awaken–up nodes N wtogether with the rate of loss p m are deduced. It is pointed out that reducing target prediction error d p can cut down the loss rate pm significantly. A mobility target prediction method based on kinematics theory is proposed, which has a good advantage in the absence of prior information of target motion. Prediction method based on particle filter is also studied, which obtains a better prediction results under a strong regularity of the target motion. The method of adjusting prediction time dynamically is researched, which achieves a higher prediction accuracy compared with the method of fixed prediction time interval. It enhances the adaptability of mobile targets markedly. Nodes wake-up mechanism based on dynamic prediction is proposed, which reduces target prediction error owing to characteristics of different movement and cuts down the rate of loss compared with linear prediction method. It can realizes nodes awaken with more measuring nodes. The energy consumption estimation formula for nodes wake-up mechanism is established. Moreover, simulation method of energy consumption based on MATLAB and OPNET is discussed.(4) The law of mobile target localization based on adaptive LSSVR regression modeling is researched. It is pointed out that whether to modeling can be decided by the containing relationship between LSSVR modeling nodes and current measuring nodes so as to reduce the number of modeling. Using Gauss-Jordan elimination method with maximal column pivoting to solve LSSVR matrix equation synchronously improves the computational efficiency of modeling. Data-centric in hourly communication mechanism and radio-style communication mechanism of nodes wake-up are studied to reduce the communication time between nodes. The rapid target localization method based on adaptive LSSVR modeling synchronously is proposed. It reduces the localization time and achieves rapid target localization in two aspects of modeling localization calculation and nodes communication. Compared to MLE, the method has higher accuracy and much litter localization time which is less than the target detection time interval and reflects a good real-time performance.(5) By synthesizing target localization, target prediction, nodes wake-up and so on, the performance of LSSVR localization method is evaluated more systematically. It is proved that it can obtain good overall localization results. Indoor positioning experiments based on the developed LSSVR rapid indoor positioning system proves the practicality and effectiveness of LSSVR localization method in the practical application environment. Moreover, the application schemes of LSSVR rapid positioning system in the manufacturing process, fire training, and shipbuilding are analyzed.Experiments of target localization based on CC2430 shows that LSSVR local modeling can improve the localization effect of different types of targets. The wake-up mechanism based on dynamic prediction can also reduce the energy consumption and loss rate. The rapid localization method of adaptive synchronous LSSVR modeling enhances real-time performance of localization. Integrated LSSVR location methods can achieve good overall localization performance. When selecting the distance value and signal strength difference as feature value respectively, the localization errors are reduced by 19%-41%,51%-58% and the localization time are 0.91-1.01s,1.02s-1.12s in modeling case, while in non-modeling case it is about 0.71-0.81s. The loss time is zero through ten experiments, while the energy consumption is close to the method of LP-MLE. Indoor positioning experiment proves the practicality and effect of LSSVR localization method in the practical application environment and achieves a good localization result. LSSVR indoor positioning method can be further applied to the manufacturing process, fire training, and shipbuilding and other areas if the nodes performance, network controllability, remote access etc are improved.