Research Of Indoor Location Algorithm Based On Multilayer Neural Network

The indoor space is more and more large and the indoor structure is more and more complex, which causes mobile terminal users unable to get the surrounding related services provided by GPS or Bei Dou Navigation Satellite System. Indoor Location Based Service(LBS) obtains widespread attention(such as a mall shopping guide services, etc). Indoor positioning technology as the underlying technology is essential for indoor LBS. The accuracy, efficiency and real-time performance of indoor positioning technology will directly affect the service quality of indoor LBS.With the improvement of the IEEE 802.11 standard, Wireless Local Area Network(WLAN) with the features of easy extensibility, easy maintenance, anti-interference performance, high mobility and flexible architecture is popular and deployed in the world. Indoor location based on WLAN with the advantages of technology of low cost, strong adaptability to environment, high location accuracy and good real-time performance has become the preferred indoor location technology, which can use received signal set to realize the localization without additional hardware facilities and provide a huge application space and technical support for indoor LBS.With the weakness of three modules(fingerprinting database establishment, APs selection, positioning prediction) in indoor location based on WLAN, this article pays attention to reducing the volatility of signal to overcome the drawbacks of three modules and improve the accuracy of the indoor location. Firstly, it analyzes the influence of the signal under the factors, such as the movement of people, the change of time. A prediction algorithm for indoor location is put forward by adopting Adaptive Generalized Regression Neural Network(IABC-GRNN), which is the basis research of indoor location. The method introduces the Improved Artificial Bee Colony Algorithm(IABC) to optimize the parameter of GRNN, which is applied to wireless indoor location for the mapping relationship between the signal characteristics and the target location and achieve a good balance between the location accuracy and adaptability. Secondly, with the problem that Access Points(APs) are not all available and the correlation between APs and subdomain is not considered, it proposes the Distributed Selection on AP(DSAP). This method divides indoor area into several sub areas and calculates the correlation between the sub area and AP nodes for selecting the AP nodes with optimal correlation as training nodes of this sub area, deep belief networks model is used for training the location model, which can effectively remove the AP nodes with noise and weak position resolution. Finally, as indoor environment is large and complex, gaining label datas is relatively difficult. According to the problem, this paper proposes a semi-supervised location algorithm based on Clustering by Fast Search and Find of Density Peaks(CFSFDP)and Extreme Learning Machine(ELM). This algorithm uses CFSFDP to cluster initial samples and label the unlabeled clustering centers, which helps to expand the initial labeled samples. Then this algorithm uses ELM to train the labeled samples and extend the labeled samples through the output threshold vector and strategy of transposition, which improves location accuracy effectively and reduces the work load of sampling.