WiFi fingerprint positioning method based on fusion of autoencoder and stacking mode

Focused on the the problem of low accuracy of floor positioning using WiFi fingerprints, a floor positioning method based on fusion of stack autoencoder and multiple models was proposed. In the process of locating through a WiFi fingerprint, the WiFi fingerprint database usually contains information of hundreds of APs, resulting in a "dimensional disaster". In order to solve this problem, for the collected WiFi fingerprint data, the feature selection is first performed by a random forest algorithm, and AP points that have little effect on the positioning result are found and deleted from the fingerprint database; secondly, an improved deep autoencoder is used to extract the features of the pre-processed WiFi fingerprint data; finally, combine multiple machine learning models such as SVM, XGBoost, ELM to analyze the positioning effect, establish a floor prediction model based on the Stacking model fusion, and further improve the accuracy of floor positioning. An experimental simulation was performed using the public data set UJIIndoorLoc, and the results show that the accuracy of the method in the test set floor positioning is as high as 95.13%, which is about 5% higher than PCA + SVM and about 3% higher than SAE + DNN.