The Prediction Of Urban Populations Based On Mobile Phone Location Data

With the increasing progress of urbanization,the pace of urban life is accelerating,and people's activities and movements are also characterized by high temporal and spatial dynamic changes.The near-real-time prediction of urban populations at the fine-grained scales can provide an important scientific basis in many fields,such as optimizing the allocation of public resources,assisting urban traffic guidance,making the early warning in urban em ergencies,as well as exploring daily life patterns of urban residents.The rapid development of information technology and mobile communication technology has resulted in a large number of individual trajectory data(such as mobile phone data,taxi data,etc.).Compared with traditional questionnaire survey data,these data have large sample sizes and strong real-time performance,especially Mobile data contains a wealth of group movement information,which can provide important data sources for studying the movement patterns of urban people,real-time crowd distribution monitoring,and travel forecasting.Based on mobile phone positioning data,this paper takes cell phone users on the spatial unit as the research object,and based on the time series analysis method,establishes a population number prediction model based on improved KNN(K-Nearest Neighbors)by considering time series autocorrelation to choose time lag reasonably and introducing linear distribution of weight distribution and Gaussian prediction functions.And use the improved KNN prediction model to conduct the urban population prediction on a fine scale and analyze the distribution of prediction errors.The proposed improved-KNN prediction model is compared with a variety of predictive models to verify its effectiveness.The results show that the improved KNN prediction model proposed in this paper can obtain better prediction results.Finally,on the basis of improved-KNN prediction model,a time-phased prediction model is established to further improve the accuracy of the population number predictions.The specific research work and features include the following aspects:Firstly,we preprocess the massive mobile phone location data by data cleaning and trajectory resampling to better mine cell phone location data information,approximate the number of cell phone users as the crowd population and analyze the spatial distribution characteristics and temporal variation rule of population.Secondly,we describe the prediction principle based on the traditional KNN model,analyzes the shortcomings in the population prediction model based on the traditional KNN algorithm and proposes corresponding improvement measures,including the reasonable determination of time lag by taking into account the time series autocorrelation,use the linear weight distribution methods to assign the weights of the historical samples and the Gaussian function is used to adjust the neighbor prediction results.So as to establish a population prediction model based on the improved KNN algorithm.Thirdly,we use the actual mobile phone signal data to calibrate the model parameters and obtain the best model parameter combination through cross-validation.Then we use the established model to realize the prediction of the number of urban mobile populations under a fine scale.Based on the prediction results,the model errors are analyzed from the perspectives of the number of population,spatial and temporal distribution,multiple time scales,and special events.The proposed model is compared with the traditional KNN prediction model,ARIMA model and historical average model to verify the validity of the model.The results show that the model proposed in this paper has the best prediction effect and achieves MAPE of2.74%.Finally,we propose the concept of time-phased modeling.We divide the day into different time intervals according to the characteristics of crowd movement and establish a time-phased KNN population prediction model,which can dynamically adjust model parameters according to time periods,so as to further improve the accuracy of near-real-time predicting of the number of population.