Research On Indoor Positioning Algorithm Based On BP Neural Network And Kalman Filter

Wireless Sensor Network(WSN)has been widely used in environment monitoring,smart home,industry control,etc.Localization is a important part in WSNs,particularly in an indoor environment where the global navigation satellite system like GPS,GLONASS,GALILEO,BDS,etc.has low positioning accuracy and high cost,because the satellite signal is blocked by roofs and walls of buildings.Therefore,in order to achieve fine-grained positioning result at low cost in the complex indoor environment.This paper uses the Received Signal Strength Indication()to locate.One of the most important challenges for indoor localization based on is that the signal is easily affected by environmental noise.To reduce the influence of noise on the positioning accuracy,two aspects are mainly studied in the following:(1)In the complex interior environment,traditional localization algorithm,such as signal attenuation model,has a low positioning accuracy.An algorithm based on BP neural network and unscented Kalman filter,named BP-UKF,is proposed in this paper.BP neural network,a fingerprint-based algorithm,is trained to estimate the distance in the first stage.In the second stage,the coordinates of nodes are initialized through the trilateration algorithm.Under the environment with unknown or timing-varying noise,the coordinates are refined through UKF for improving the positioning accuracy.Experimental results show that the position estimation error of BP-UKF algorithm is smaller than BP-EKF algorithm,especially for the tradition localization algorithm.(2)UKF does not have the property of online estimation of noise statistics,which makes it difficult to improve the positioning accuracy of BP-UKF algorithm no longer in an indoor complex environment.In this paper,a novel adaptive unscented Kalman filter(AUKF)is proposed based on the existing adaptive Kalman algorithm with maximum likelihood estimation by improving the noise statistical estimator.An algorithm based on BP neural network and a novel AUKF,named BP-AUKF,is proposed.Proposed BP-AUKF algorithm can perceive and update noise parameters online,and improve the positioning accuracy.The effectiveness and accuracy of BP-AUKF algorithm are verified via a large number of simulation.And the positioning estimation error of BP-AUKF algorithm is also small in a real environment,compared with the existing filter algorithms.