Locality Preserving Canonical Correlation Analysis Distributed Localization Algorithm for Wireless Sensor Networks
International Journal of Multimedia and Ubiquitous Engineering
Localization is essential for wireless sensor networks. The state-of-the-art methods mainly adopt low accurate signal strength to perform localization, which are suffering from low localization accuracy and high variance. The machine learning methods are introduced to confront the low data quality challenges and provide considerable localization accuracy and other advantages. However, these series of methods also bear some drawbacks such as high training cost and high energy consumption. To
... consumption. To this end, learning from our previous algorithm LE-LPCCA (Location Estimation-Locality Preserving-Canonical Correlation Analysis), we proposed an improved version, called, LE-DLPCCA (LE-Distributed-LPCCA), which greatly reduces the training cost and energy consumption. Specifically, LE-DLPCCA employs a clustering algorithm based on energy equilibrium. The training process, which maps the signal space into physical space, is conducted in a distributed manner for each cluster. Then, in the positioning phase, the unknown node estimates the distances from the most similar anchor nodes through the mapping and perform the localization of the unknown nodes through the maximum likelihood method. Demonstrated by multiple simulations, LE-DLPCCA algorithm is in high accuracy, fast localization model efficiency and low average energy consumption. Centroid algorithm , Convex programming algorithm, DV-Hop algorithm , APIT algorithm  and so on. The above localization algorithm has been widely used. Due to the hardware equipment, the external environment and other factors, they result in a larger error, many algorithms have been improved. ZANCA et al. present an indoor localization algorithm  based on RSSI in wireless sensor networks, which needs no additional hardware equipment, low power consumption and low cost. However, it lacks of stability and the localization accuracy is poor. LEE et al. propose a wireless sensor network localization algorithm  based on improved AOA algorithm in Ad-Hoc network, which requires additional hardware, vulnerable to the impact of the environment and the result is rather unstable. CHEN et al. use global beacon nodes of the average hop distance  to calculate the distance between unknown nodes and beacon nodes, and adopt hyperbolic localization method to solve for the coordinate of the unknown node. Xu et al. put forward a APIT based centroid localization algorithm  to overcoming the localization error caused by the uneven and sparse beacon nodes, which improves localization accuracy. But this algorithm applies only to sparse anchor nodes in wireless sensor network. It is not suitable to the general wireless sensor networks. Overall, the above localization algorithms are mainly unstable, low localization accuracy and high energy consumption. In order to solve these problems, researchers have provided a lot of improvement strategies and algorithms. In recent years, modeling and localization by machine learning has become one of the research focuses [1-2], which regard the essence of the wireless sensor localization process as a machine learning process. Researchers In-depth mine the available data in wireless sensor networks such as signal strength, physical coordinates, network topology information. Then we establish a localization model for wireless sensor networks to estimate the physical coordinates of the unknown nodes. In this paper, we propose an improved machine learning localization algorithm and construct a model to improve the positioning accuracy, enhance the speed of modeling and reduce the average energy consumption. The proposal utilizes the clustering algorithm to divide the monitoring area into many subdomains and establish the mapping between the signal space and physical space. In addition, we adopt the maximum likelihood method to estimate the physical coordinates of the unknown nodes. Finally, the performance of the algorithm is verified and compared with other methods. The simulation results show that the proposed algorithm outperforms the other two methods in terms of accuracy, speed of modeling and the energy consumption. LE-DLPCCA makes the following contributions: a) The algorithm can provide a more accurate and stable method to estimate the physical coordinates of the unknown nodes. b) The cluster head nodes of subdomain perceive the invasion of unknown nodes, which awaken the other cluster nodes. So it can reduce the energy consumption of sensor nodes. c) The algorithm can establish the whole model by combination of sub models, so we can increase the speed of modeling. Outline Section II briefly introduces the related work of machine learning localization algorithm. In Section III, the LE-DLPCCA localization algorithm is described in details. In Section IV, exemplifies it with the simulation to illustrate the efficiency of this algorithm and make the conclusion in the Section V.