Research On Indoor Localization System Based On Weighted Hybrid Regression And Improved Algorithm Performance

With the continuous advancement of technology,GPS,as the most common navigation and positioning system,has spread all over the world,and it is developing in the direction of higher accuracy,increasingly miniaturized equipment,and coexistence of multiple positioning technologies.Outdoor positioning technology has been very mature,but the development of indoor positioning technology is relatively backward.Due to the severe attenuation and multipath effects of the signal in the indoor environment,general outdoor positioning technology and hardware facilities cannot work effectively in the indoor environment.So far,there has not been a relatively mature indoor positioning technology applied to people's daily life.The realization of the indoor positioning system requires the establishment of a complete set of positioning facilities,including fixed hardware facilities installed in a fixed location to send positioning signals and mobile equipment to receive signals.The deployment area of the signal transmission facility is relatively large,and the cost investment is also relatively high.Therefore,this thesis chooses to use the widely deployed wireless device(Wi Fi router)to conduct the research of indoor positioning technology.In order to improve the accuracy of indoor positioning algorithm,this thesis mainly does the following work:?.In order to improve the accuracy of the positioning algorithm and make the algorithm more adaptable to the positioning environment,this thesis proposes the PL-SKR weighted hybrid regression algorithm.Unlike the traditional use of a single regression algorithm to achieve the regression positioning of the entire region,the innovation of the PL-SKR algorithm lies in the parameter optimization and model training of the regression algorithm in different regions.And adopt a weighted way to realize the hybrid regression of SVR and KNR regression algorithms.First,in the offline stage,the clustering results of the K-Means algorithm are used to divide the positioning area into four sub-areas,and then the regression learning and parameter optimization of the KNN and SVM algorithms are performed in each sub-area.In the online phase,the PL-SKR algorithm uses two regression models to perform positioning tests on the training set samples in the category to which the node to be tested belongs,and the probability of small positioning error of the two regression models is calculated respectively.Then,we assign large weights to the regression model with a small cumulative probability of positioning error,and select this regression model for position prediction.After experimental simulation tests,the average positioning error of the PL-SKR algorithm is 0.76 m.Compared with the positioning algorithm using only SVM or KNN,the average positioning error of the hybrid regression algorithm is reduced by 1.72m?1.99 m.?.The PL-SKR algorithm has made certain improvements in reducing indoor positioning errors.However,there are many reasons for the increase in positioning error.In order to reduce the positioning error caused by different factors in the offline and online phases of the indoor positioning process and improve the efficiency of the algorithm,this thesis proposes four algorithms.Aiming at the problem of incomplete data noise filtering,this thesis proposes a data multiple filtering algorithm(Mul?data).The algorithm filters the positioning data by setting thresholds for the range of signal strength values and analyzing the mean and variance of the signal strength vectors.The experimental results show that the signal feature vector processed by the Mul?data algorithm has 26 dimensions less than the feature vector processed by the Gaussian filter.The reduction of the dimension of the signal feature vector helps to reduce the computational complexity of the positioning process.Aiming at the problem of algorithm operation efficiency,this thesis proposes a data compression algorithm(Re PCA).The algorithm uses the Relief F feature selection algorithm to perform feature selection on the fingerprint database data.Then the PCA algorithm is performed on the data to reduce the dimensionality.So as to realize the data compression function.The experimental results show that,before the PL-SKR algorithm parameter adjustment and model training,we first process the data through the Re PCA algorithm,which can reduce the time consumption by about 64.7%.Aiming at the problem of wireless signal changes in the measurement environment,this thesis proposes a wireless signal monitoring algorithm(W-Monitor).The algorithm compares the collected signal feature vector with the fingerprint database by traversing the fingerprint database,thereby calculating the matching degree between the feature vector and the fingerprint database.We use this algorithm in the positioning process of the SVR algorithm.Through experimental simulation,we found that if we use the W-Monitor algorithm to calculate the matching degree between the signal feature vector and the fingerprint database before the positioning,if the matching degree is less than 80%,we do not continue this positioning.In this way,a positioning result with large errors can be avoided.In order to estimate the size of the positioning error,this thesis proposes an error prediction algorithm(self-inspection).The algorithm uses the same type of data samples to test the positioning error and sets the conditions for the allowable value of the positioning error,so as to control the positioning error within a certain range.We use this algorithm in the positioning process of the SVR regression algorithm.Through experimental simulation,we found that if the self-inspection algorithm is added to the online positioning process of the SVR algorithm,the average positioning error of the positioning result can be reduced by about 1.2m.According to the algorithm design needs,we can selectively apply the four algorithms to the indoor positioning process.