類神經網路應用於室內通道定位之研究

The Research of Artificial Neural Network for Indoor Walkway Location

吳鴻志(Hung-Chih Wu) ； 張振明(Chen-Ming Chang)

《2008資通技術管理與應用會議》 2008 (2008/12) Pp. 617-628

摘要

近年來有許多的技術陸續以使用者定位技術來發展，而無線感測器網路就是其中一種，無線感測器網路(Wireless Sensor Networks, WSN)具備低成本、低耗電、小體積、容易佈建，並可程式化、可動態組成等特性。本論文是根據實地量測資料以徑向基－廣義迴歸類神經網路(GRNN)及倒傳遞類神經網路(BPN)建立室內通道電場強度並結合電腦作精密定位的控制，行動端將測量各基地台的訊號強度值(RSSI)，並使用類神經網路的演算法和訊號強度(RSSI)進行訊號偏移分析，而在克服(RSSI)訊號變化上本論文模擬狀況為1個基地台訊號變化為5%時，徑向基－廣義迴歸類神經網路(GRNN)均方根誤差是0.40，命中率約85%，倒傳遞類神經網路(BPN)經訓練收斂至10^(-25)，均方根誤差是1.26，命中率約43%。實驗結果顯示徑向基－廣義迴歸類神經網路(GRNN)來進行室內定位工作，相較於倒傳遞類神經網路(BPN)，可於(WSN)訊號發生偏移時可獲得良好定位效果，兼具定位系統可靠性與降低系統建置所需的人力與時間成本。

關鍵字

無線感測網路 ； 訊號強度 ； 徑向基類神經網路 ； 倒傳遞類神經網路

並列摘要

In recent years, numerous technologies have been successively developing based on positioning technology, and wireless sensor network (WSN) is one of them. Wireless Sensor Network features low cost, low power consumption, compact, easy deployment, as well as programmable and capability of dynamic composition etc.
This study has built an indoor channel of electric field intensity integrated with computers for accurate positioning control with Generalized Recurrent Neural Network (GRNN) and Back-Propagation Neural Network (BPN) according to the practically measured data. The mobile-end will collect the Received Signal Strength Indicator (RSSI) of each base station for analyzing the signals deviation. On the other hand, the study simulated in overcoming the RSSI signals variation as: when a given base station with its signals variation at 5%, the Root Mean Square Error of GRNN is 0.40, the hit rate is approx. 85%; while BPN converged to 10-25, the Root Mean Square Error is 1.26, and hit rate approx. at 43%. Results showed that by using GRNN, instead of BPN, the better positioning effect can be obtained when WSN signals shifted, which will enhance the reliability of the positioning system and reduce the manpower and time cost as required for system deployment.

並列關鍵字

Wireless Sensor Network WSN ； Received Signal Strength Indicator RSSI ； Generalized Recurrent Neural Network GRNN ； Back-Propagation Neural Network BPN