Detection Methods Of Residents' Activity-travel Characteristics Based On GPS Trajectory

With the rapid development of the urbanization and economy in our country,traffic congestion has become the most important "urban disease" in many cities.Along with traffic congestion,resource shortage and environmental pollution and other problems are getting more serious.To ease the growing traffic congestion in the city,transportation management departments made a series of transportation demand management policies to adjust the temporal and spatial distribution of transportation demand.The establishment of these policies depends on the understanding of residents' travel patterns and regularities.A travel surveys is an important tool to obtain residents' activity-travel patterns and provide a foundation of constructing residents' travel patterns.However,in the traditional travel surveys,respondents undertake a heavy burden,and the accuracy of the travel characteristics is low,and the routes of respondents cannot be easily obtained.As the positioning technology progresses rapidly,GPS based travel surveys provides a perspective to solve these issues.GPS based travel surveys can collect accurate location information,but cannot obtain travel characteristics including trip ends,travel modes and trip purposes.Therefore,it is very meaningful to explore GPS based travel surveys and detect trip ends,travel modes,and trip purposes based on the GPS data collected.Under the practical background,in the light of the deficiency included in existing studies on GPS based travel surveys and data mining on the basis of GPS data,and based on the high marketing penetration rate of smartphones and the popularity of Internet in China,we explicitly analyze and discuss the feasibility of travel surveys based on smartphones and Internet technologies;from the perspective of three dimensions consisting of trip ends,travel modes,and trip purposes,we comprehensively compare the advantages and disadvantages of existing methods on travel characteristics mining;based on the aforementioned analysis,we discuss the methods to improve the detection accuracy of travel characteristics.This study can prompt the application of time geography on the field of travel behavior analysis and prediction,enrich and extend the theoretical and methodological system,promote the application of the theory and method of data mining on the field of activity-travel behavior surveys,enrich the data source of the analysis on activity-travel behavior,and extend the collection methods of accurate travel characteristics.From the perspective of data collection and data mining,this study employs knowledge of time geography,transportation engineering,behavior science,management science,statistics,and information science to analyze the application of GPS positioning technologies,Internet technologies,and data mining methods on residents' travel survey,and its main contents include(1)Travel surveys based on smartphones and Internet survey technologies are employed.First,based on GPS data collection used to explore travel characteristics,the development requirements of positioning application on smartphones are proposed.Second,according to the object of travel behavior modelling and undertaking the travel survey based on smartphones and Internet survey technologies,we design an online questionnaire aiming to collect individual and household socio-economical attributes of respondents.Third,according to the principle of minimizing the burden on respondents and maximizing the accuracy of travel characteristics,we propose travel surveys based on smartphones and Internet survey technologies.(2)Taking GPS data as input,we propose a method of detecting trip ends based on grid search.Taking the GPS data collected in the travel survey based on smartphones and Internet survey technologies as the data source,we analyze the completeness of the data.Based on the analysis,we discriminate the two scenarios of GPS signal loss and normal recording,and separately propose the trip end detection method based on the grid search of parameters.Based on the validated travel characteristics in travel surveys,we then discuss the method of evaluating the parameter combination.Under the best parameter combination,the recall rate achieves 96.02%,while the error rate is only 4.75%.Results indicate that travel surveys based on smartphones can complement traditional travel surveys and that the travel surveys can provide a theoretical and methodological guidance for further undertaking large-scale GPS based travel surveys.(3)A two-stage method on travel mode detection based on GPS travel surveys is proposed.Based on the characteristics incorporated in each travel mode,we propose a two-stage method to detect travel modes.In the first stage,we employ grid search based on GIS spatial analysis to detect subway modes;in the second stage,we employ Gaussian process classifier to detect travel modes other than subway.Before using Gaussian process classifier,we employ the sequential forward selection method to select the features used to classify travel modes.By comparing the derived and the validated travel characteristics,the algorithm correctly flags over 97% of subway samples and 95.05% of the total samples.Results indicate that the data mining method is a useful tool to promote a large-scale application of GPS based travel surveys.(4)Multiple scenarios on trip purpose detection are constructed,and the method on trip purpose detection based on GPS travel surveys is proposed.Based on existing studies,we propose four scenarios to detect trip purposes.In the light of the incomplete features employed in existing studies,we discuss the feasibility and necessity of employing the activity duration and the time when activities start in the trip purpose detection.Under each scenario,taking the feature set including the activity duration and the time when activities start as the data source,we employ Artificial Neural Networks(ANNs)and Particle Swarm Optimization(PSO)to detect trip purposes.By compared the derived and validated travel characteristics,the algorithm achieves an accuracy of 97.22% for the training dataset and 96.53% for the test dataset.Results show that features associated with activities(including the activity duration and the time when activities start)have a positive effect on the trip purpose detection.This study comprehensively analyzes the method of collecting GPS data and detecting travel characteristics,and has an important theoretical and practical meanings on the improvement of residents' travel surveys.Compared with existing studies,the originality of this study includes(1)We propose the travel survey based on smartphones and Internet survey technologies.Different from existing studies,from the perspective of decreasing the burden on respondents and improving the accuracy of the travel characteristics,this study constructs an online platform including the individual socio-demographic attributes,GPS data,and the mining of GPS data,and discusses the respondents' travel characteristics validation in the case of the surveyor intervention.(2)We propose the method on trip purpose detection based on GPS travel surveys.Based on the application of grid search on trip end detection,we taken the overlapped road length as an important parameter to detect trip ends for the trip purpose of “picking up/dropping off someone”.The method of grid search can effectively avoid the deficiency of the trip end detection method due to the empirical experience included in existing studies,and thus improve the accuracy of trip end detection.(3)We propose the method on travel mode detection based on GPS travel surveys.We split the detection of subway and other travel modes into two stages for the first time,and separately employ GIS spatial analysis and Gaussian process classifier to detect travel modes.This two-stage method explicitly consider the significant feature difference included in the subway and other travel modes,and provides a new perspective for improving the accuracy of travel mode detection.(4)We propose the method on trip purpose detection based on GPS travel surveys.By comprehensively considering different scenarios included in existing studies,we construct multiple scenarios for trip purpose detection.Under each scenario,we employ ANNs and PSO to detect trip purposes.ANNs can comprehensively describe the complex nonlinear relationships between trip purposes and the corresponding features,and PSO can find the global optimal for the parameter optimization with a relatively large probability.The application of these methods provides a preferable foundation for improving the detection accuracy of trip purposes.The contents,methods,and conclusions of this study is a further exploration and complementation of existing studies on travel surveys based on GPS,provides a theoretical guidance for further studies on the aspect,and provides a theoretical basis and decision foundation for the improvement of collection methods on residents' travel characteristics.