Research On Location Data Attribute Mining And Location Prediction Technology

With the rapid development of modernization,it is easy to collect global navigation satellite system(GNSS)trajectory data during a human journey.This type of data is collected and recorded at different time points,and reflects the state of moving objects or a certain phenomenon over time.Through the location data to discover the law of human life movement and human activity pattern.This article discusses issues such as identity recognition,traffic mode identification,behavior pattern recognition,and location prediction based on GNSS data.From the above four aspects to do the research on location data attribute mining and location prediction technology.Firstly,identify the user’s identity,then infer the user’s traffic mode(eg,walk,car,train,etc.),and then further identify the behavior pattern of the transportation mode(eg,turn left,turn right,go straight,etc.).The data source for identification is that the wechat applet which can directly collect the recorded acceleration and compass direction data.Secondly,because the traffic modes and behavior patterns are strongly affected by the geographical environment,geographic layer information(eg,buildings,roads,water,etc.)will enhance the identification of transportation modes and behavior patterns.Finally,the user’s location is highly related to the geographic layer information.Therefore,the geographic layer information can assist GNSS data for location prediction.The geographic layer attribute information can be downloaded from the OpenStreetMap(OSM).There are six types of layer information in total.The geographic layer information is added to be able to know the geographic environment information around the GNSS point.The implementation process is as follows.Firstly,at a specific GNSS point,at a specific GNSS point,the surrounding area is sliced into grids uniformly.Secondly,the probability of grid centers belonging to six different geographic layers is calculated according to whether the centers of these grids are inside or outside the smallest rectangle containing polygons indicating different geographical objects.Finally,a six-dimensional geographic attribute information probability matrix of GNSS points is obtained through comprehensive processing.The six-dimensional probability matrix is processed and compressed as a geographic information vector through a convolutional neural network(CNN).The latter is then combined with kinematic indicators(such as speed,acceleration,and direction of motion from GNSS data)and serially input into a long-short-term memory(LSTM)network to predict traffic modes and behavior patterns.The experimental results verify that the geographic layer information indeed enhances the performance of the two recognition tasks.The CNN+LSTM framework retains the functions of CNN and LSTM,and is superior to traditional machine learning algorithms,the accuracy of the CNN+LSTM model is more than 6%higher than several algorithms of machine learning.The position prediction part uses a hybrid network of a three-dimensional convolutional neural network(3D-CNN)and a LSTM network.3D-CNN is used to process geographic layer information,and then use the output and kinematic information extracted from the GNSS data as input to the LSTM.We can obtain the probability value of each mesh through 3D-CNN and LSTM models,then use this probability value to update the weight of the particle filter,and finally obtain the specific position of the user.The root mean square error(RMSE)value is less than half of the value of the original particle filter.Experimental results show that this is superior to traditional solutions.