- 學位論文
在ZigBee室內定位系統中以倒傳遞類神經網路提升區域定位準確性之研究
Research on Back Propagation Neural Network to Enhance Area Location Accuracy in ZigBee Indoor Positioning Systems
摘要
本論文針對ZigBee室內定位系統中，我們提出利用倒傳遞類神經網路提升區域定位的準確性。在無線定位系統中，訊號傳遞過程易受環境影響而造成訊號失真或衰減，這將會影響定位的準確性。為了降低因訊號不良而導致的定位誤差，因此本文採用樣本比對(Pattern matching)的方式建立環境模型，此種方式無需測量待定位目標物體與定位參考點之間的距離關係，而是比對待定位目標物體所蒐集的訊號強度特徵與該環境定位資料庫中各個訓練位置的特徵，此方式可以有效降低因為訊號不穩定所造成的定位誤差。透過環境模型將可利用定位演算法建立該環境的定位模型，並利用此定位模型定位該環境上的目標物體。本文將使用K-最近鄰居平均(K-Nearest Neighbor Average, KNN-AVG)演算法與倒傳遞類神經網路(Back Propagation Neural Network, BPNN)進行定位準確性的比較。在實驗方面，我們利用ZigBee建立無線感測網路(Wireless Sensor Network, WSN)作為定位環境，其定位參考點也可作為感測節點來使用，其目的為蒐集周遭環境的物理變化量並儲存於後端資料庫進行分析。而在定位方面，分別使用不同的環境進行實驗，主要分為較空曠環境與較複雜環境，並針對不同地圖大小進行各演算法的比較。由實驗結果得知，由於倒傳遞類神經網路具有學習能力，只要適當的調整倒傳遞類神經網路的學習參數，比起K-最近鄰居平均演算法更能提升區域定位準確率。
並列摘要
In this thesis, the back-propagation neural network is used to enhance area location accuracy in ZigBee indoor positioning systems. In a wireless positioning system, signals would be distorted and attenuated by the environment during the signal transmission. In the case, the distorted signals would degrade positioning accuracy. In order to reduce positioning errors due to bad signals, this thesis uses a pattern matching approach to build environmental model. The pattern matching approach does not need to measure the distance relationship between the target object and the positioning reference points. We only need to compare the collected signal strength characteristics from the target object with the collected database features of the various training locations in the environment. Based on this, we can reduce the position error due to the unstable signals. According to the environmental model, we can use the position algorithms to locate the target in the environment. We would compare the position performance of the K-nearest neighbor average (KNN-AVG) with back-propagation neural network (BPNN) in terms of positioning accuracy. In the experimental environment, we would use ZigBee platform to build a Wireless Sensor Network (WSN) for area location. Its reference points cannot only be for the use of location but also be as a sensor node. It can collect the changes of physical in the surrounding environment and be stored in the back-end database for analysis. In positioning, we locate the target in various environments which divide into the open environment and the complex environment. We compare the location algorithms in different size of the map. From the experimental results, since the back-propagation neural network is capable of learning, we could adjust the learning parameters of the back-propagation neural network appropriately to increase positioning accuracy. Compared to the K-nearest neighbor average algorithm, the back-propagation neural network could enhance area location accuracy and have better performance.