| As a new type of wireless ad hoc network, wireless sensor networks have wide applications including industry, agriculture, environmental observation, military, and collecting information in disaster prone areas. Knowledge of nodes location is an essential requirement for many applications. Without location information of sensor nodes, the sensed data or the detection of events is nonsense. However, because each sensor with a GPS device is expensive in terms of cost and energy consumption, it is infeasible due to various resource constraints such as miniature size, low-complexity, limited battery power. Therefore, self-localization for wireless sensor networks has been presented and studied. Based on the key characters of wireless sensor networks and the limitations of current research, this thesis focuses on localization issues and related problems from stationary, beacon nodes moving and mobile wireless sensor networks.This thesis firstly studies the localization problems in static wireless sensor networks. In incremental or multi-hop range-based localization algorithms, the ranging error or localization error can be easily accumulated, and this will affect the localization accuracy of successor nodes. Some range-free centralized localization algorithms can obtain high localization accuracy only using network connectivity information and do not need to consider the ranging error. However, centralized localization needs to collect the information of the entire network to base station, thus the communication cost is high and it consumes too much energy. Aiming at these drawbacks, we present a semi-centralized localization algorithm based on support vector regression. The base node collects all connectivity information between beacon nodes and applies these collected information as training samples to run the training procedure by using support vector regression method. As a result, a regression function can be derived and be distributed to all sensors in the network. Then, normal nodes can perform the estimation of locations using the function based on the connectivity to all beacon nodes. In order to increase the number of training samples, the normal nodes having minimum three anchor nodes as neighbors are upgraded to anchor nodes. Least-square method with RSSI-based ditance measurements are applied to those normal nodes so that they can locate their positions. Our algorithm is composed of range-based method and range-free method. But it can avoid collecting global network information and so reduce the accumulation of the ranging errors.For resource-limited and self-organized wireless sensor networks, centralized localization methods are impractical because of the high communication overhead and bad scalability. We turn the localization problem into an unconstrained optimization problem and propose a distributed localization algorithm based on hybrid taboo search. In order to alleviate the effect of ranging error and get a good interval to generate neighboring solutions, a selection operator is presented to select several suitable anchor nodes to take part in the localization procedure. Subsequently, information of anchor nodes is used to obtain initial estimate of location of nodes. Finally, an optimization procedure is performed by using a hybrid method composed of taboo search and simulated annealing. This improves the accuracy of the estimation of the initial location. The simulation shows that the proposed algorithm has a good accuracy on localization when noise factors and the number of anchor nodes are different. Instead of having some anchor nodes in the most of existing current localization methods, mobile anchor node-assisted localization algorithm only needs one or several anchor nodes which traverse the deployment area to help normal nodes locate. Such architecture significantly reduces the cost of networks. However, the path of mobile anchor node has a direct impact on the performance of these approaches. And not too much research work has been done on this. Several static path planning methods for mobile anchor node have been presented based on a certain curve. Such static methods cannot make use of the real-time information in localization process, and are unsuitable for irregular network topology. To improve the static methods, we propose a novel heuristic dynamic path planning method with a better flexibility. It is the first time that directional antenna technology is used to solve the path planning problem. By communicating with neighborhood normal nodes, mobile anchor node configured with directional antenna array can detect the number of distributed normal nodes and the number of beacon which each normal node has received from it. Based on the knowledge, the mobile anchor node can make on-line decision for its moving. In order to reduce the communication overhead, the mobile anchor node make one decision and move twice. The simulation results show that our dynamic planning method can effectively avoid covering the areas where no sensor nodes are deployed. This reduces the distance of the moving path and the amount of messages transmitted by mobile anchor. The simulations also indicate that our method provides good localization coverage.In mobile wireless sensor networks, the network topology and the connectivity between all the nodes continually change because of the mobility of normal nodes and anchor nodes. Therefore, this brings more complicated and difficult problems in localization. In order to enhance the convergence speed of localization algorithm and improve the localization accuracy and timeliness, we present a Monte-Carlo localization algorithm based on dynamic grid division. Firstly, we construct a farthest distance selection model. The model can save the system energy by reducing reduce the number of anchors taking part in localization, and keep a good localization accuracy. Then, these selected anchor nodes are used to create the region to draw samples from. In addition, a probabilistic approach is introduced to calculate the maximum sampling number. Finally, sampling, filtering and location estimation are executed. Unlike the existing Monte-Carlo algorithms, a mobility model with error compensation is applied to filter the samples. This model can reduce the computing overhead for a large number of loop computing. The simulation demonstrates that our algorithm reduces the sampling numbers, energy consumption and the processing time and still has a good localization accuracy.Finally, for better research the characters of wirelss sensor networks and analysis the performance of the proposed algorithms in real system, we implemented a prototype system for wireless sensor networks. The nodes of the system are designed by ourselves with ZigBee protocal to communicate among the nodes. In our system, we write a program to manage and supervise the information from nodes. Furthermore, one proposed localization algorithm is evaluated on the prototype system based on RSSI ranging technology.
|
PDF Full Text Request