Research On Indoor Intelligent Localization Based On Broad Learning System Utilizing Ensemble Strategy

Indoor localization technology based on Wi Fi fingerprints has been widely studied because of the popularity of hardware facilities.In the fingerprint-based model,the online and space complexity of the method based on similarity comparison is too high,and increases linearly with the growth of scene and fingerprint database,while the offline training of the method based on neural network is too time-consuming.The quality of service provided by indoor positioning largely depends on the positioning accuracy and time cost.Therefore,how to ensure the positioning accuracy of the model based on reducing the online and offline complexity has become a problem to be solved.This thesis proposes a stacking Ensemble Broad Learning Localization system using Channel State Information(CSI)as a fingerprint,which is termed Ensem Loca.Broad Learning System(BLS)is a concise and efficient neural network model.As a base learner,it not only does not need to store all fingerprints but also has the advantage of time complexity.The application of sparse representation further improves its ability of feature extraction.The Bagging framework allows multiple base learners to be built in parallel,so the Ensemloca model has similar time complexity to the base learner BLS.At the same time,the Bootstrap method and feature sampling strategy are used to generate sub-sample sets,and then a base learner with unique recognition ability can be constructed.The alienated base learning device deeply excavates the signal characteristics of different positions and realizes the trestle generalization through the nonlinear combiner,so that the positioning accuracy of the Ensemloca model can be greatly improved while maintaining the complexity of BLS.The experimental results show that Ensemloca has higher accuracy than several machine-learning algorithms in both Line of Sight(LOS)and non Line of Sight(NLOS)environments,and is even stronger than deep neural networks characterized by accuracy.Because there is no need to store fingerprints,it also has an advantage in space complexity.At the same time,it has the same theoretical complexity as BLS,which greatly reduces the time required for off-line training.In the experiment,Ensemloca spends the shortest time in the offline phase except for BLS,which is enough to prove its advantage of offline time complexity.The online phase only takes the calculation time related to the size of BLS neurons,which is independent of the size of the fingerprint database,which solves the problem of linear growth of time.