基于支持向量机原理的无线传感器网络定位方法研究

In the Wireless Sensor Network (WSN), the event location or the node location is one of the basic functions. Localizatlon in the Wireless Sensor Network is a key technology and research hotspot. This project proposed to study a new localization method based on support vector machine (SVM). This algorithm takes the localization problem as a machine learning regression modeling. A model will be built based on relationships between network connectivity information, received signal strength and node location positioning by the support vector regression tool LS-SVM. By this algorithm, node localization information can be obtained automatically, efficiently, and accurately. Meanwhile, the localization error is within 3% and the coverage holes and border problem is solved to some extent. Focused on the requirements of low cost and low power in wireless sensor network, this project will study a new reduction strategy for large-scale training sample set according to analyzing on the structure of the training sample set to reduce energy consumption that caused by SVM training process. Promisingly, it can shorten the time of the whole training for 50%. And then consumption of decision node.Promisingly, reduction rate of decision time will reach 50%. Finally, this project will study a novel SVM parameters optimization algorithm to improve the localization accuracy. The localization method proposed by this project dose not need additional hardware equipment and meets the requirements of low power.

定位问题是无线传感器网络中的基础问题，也是当前的研究热点。针对此问题，本项目提出研究一种新的基于支持向量机（SVM）的节点定位算法。该算法将定位问题看作一个机器学习回归建模问题，通过支持向量回归工具 LS-SVM 建立网络连通性信息和接收信号强度指示信号与节点位置的定位模型。通过该算法可以实现自动、高效、准确的节点定位，定位误差控制为3%以内，同时在一定程度上解决覆盖漏洞和边缘问题。针对于无线传感器网络低成本、低功耗的要求，研究一种新的支持向量机的训练样本缩减算法来降低由支持向量机学习过程而带来的训练节点的能量消耗，预计该算法可使训练时间缩短50%左右；通过研究一种新的支持向量机快速决策算法，提高决策节点的决策速度进而降低决策节点的能量损耗，预计该算法使决策时间缩减率达到50%以上。最后研究一种新的支持向量机模型参数优化算法提高定位准确度。该方法既不用添加额外的硬件又兼顾了低功耗的要求。

定位问题是无线传感器网络中的关键技术，也是当前的研究热点。针对此问题，本项目提出一种新的基于支持向量机（SVM）的节点定位算法。该算法将定位问题看作一个机器学习建模问题，通过支持向量机工具 LS-SVM 建立网络连通性信息和接收信号强度指示信号与节点位置的定位模型。通过该算法可以实现自动、高效、准确的节点定位，定位误差控制为5%以内，同时在一定程度上解决覆盖漏洞和边缘问题。针对于无线传感器网络低成本、低功耗的要求，提出基于点集的支持向量机训练样本缩减算法来降低由支持向量机学习过程而带来的训练节点的能量消耗，通过仿真实验证明该算法可使训练时间缩短50%。在大规模的无线传感器网络中，每个节点需要经过多个支持向量机模型分类得到最终定位结果，这致使定位时间过长，算法复杂度也造成节点能耗的大量消耗。针对此问题，本项目提出了基于空间相似性的支持向量机快速决策算法，提高决策节点的决策速度进而降低决策节点的能量损耗，实验结果表明该算法在保证精度损失的情况下使决策时间缩减48%。支持向量机的分类结果决定着最终定位精度。支持向量机的性能依赖于模型参数的选择，然而支持向量机模型参数的选择在很大程度上依靠经验值。针对这一问题，本项目提出一种基于免疫克隆算法的支持向量机模型参数优化算法，通过该算法可以自动获得最佳的模型参数，实验证明该算法可以将定位精度提高0.2M。