中文摘要：

基于WLAN（Wireless Local Area Networks）的无线定位是移动互联网领域的重要研究内容之一.其中,指纹定位方法已成为主流,此类方法的特点之一在于需要离线训练数据与在线测试数据具有严格的一致性.但在实际环境中,无线信号数据波动较大,存在显著的时效性问题.这导致一定时间后,定位模型的预测精度不断下降.文中提出一种具有时效机制的增量式定位方法（Timeliness Managing Extreme Learning Machine,TMELM）,一方面满足实际系统的应用需求,可随时加入新的训练数据进行在线增量式学习,另一方面融入时效机制,以最大化新增训练数据对定位模型的贡献,保持定位模型的精度.实验表明,在实际WLAN定位数据集上,文中方法相比于传统的几种增量式学习方法,具有明显的时效优势,能获得更好的定位精度.

英文摘要：

WLAN based localization is becoming an important research topic in mobile Internet field, and the fingerprint methods are the trend. One feature of those methods is that the accuracy is determined by the identification of the training and the testing dataset. However, in the real world, the WiFi data is changing as the time passed, which leads to the reduction of the predic- tion accuracy. We propose TMELM to handle this problem. On one hand, it meets the require- ment of the real application, allowing appending new incremental training data to the prediction model. On the other hand, timeliness management is designed to maximize the contribution of the new training data to the model. The experiments show that compared with several traditional incremental methods, our proposed TMELM performs better on the real WiFi dataset.