OBE 이력자료를 이용한 이상치 제거 범위 산정

Abstract

도시교통정보시스템(Urban Traffic Information System, UTIS)과 지능형교통체계(Intelligent Transportation System, ITS)를 연계하여 정확한 교통정보를 수집 및 분석하여 다양한 매체를 통해 정보를 제공할 수 있는 교통체계를 구축하여 교통정보의 질적 향상 및 양질의 교통정보를 이용자에게 제공하여 도시지역의 통행을 향상하고 있다. 그러나 UTIS의 경우, 개인 및 영업용 차량의 업무 및 돌발 상황으로 인해 노변에 주정차 하거나, 무선통신 장치의 이상으로 인한 데이터 중복 등 여러 가지 요인으로 인하여 교통정보 신뢰도를 저하하는 현상이 발생하고 있으며 이를 해결하기 위해서 이상치를 제거하는 연구가 진행되어 왔다. 하지만 이상치 제거 시 사용되는 사분위편차와 관리도모형에서 산정된 범위 값을 일률적으로 적용하여 이상치를 제거하고 있으며 이는 5분마다 수집된 정보를 가공하여 제공하는 시스템의 특성상 영향을 미칠 것으로 판단된다. 따라서 이를 각각의 범위 값을 적용하여 이상치 제거 시 어떠한 영향을 미치는지 비교분석 할 필요가 있다. 따라서, 본 연구에서는 OBE 구간정보제공 신뢰성 향상을 위하여 실시간 OBE 교통정보의 개별차량 자료를 5분 단위로 가공 시 이상치를 제거하기 위해, 실시간 OBE 자료의 이상치 제거 방법은 통계 분석의 사분위편차와 관리도 모형으로 사용하였으며, 5분 집계자료의 수집건수에 따른 적합한 방법과 이상치 제거 범위를 제시하고자 한다. 안산시 시내의 단일교차로 1개 구간으로 설정하였으며, 자료는 한양대학교 교통정보센터에서 수집된 2011년 9월 01일부터 9월 30일까지 1개월 동안의 OBE 원시자료를 사용하였다. 과거이력 데이터의 5분 단위 평균속도를 생성한 다음 실시간 5분 데이터의 평균속도와 중위 값을 구한 다음 이를 과거 평균속도를 기준으로 실시간 평균속도가 과거 평균속도와 오차가 적으면 관리도 모형을 적용하며, 중위 값과 과거 평균속도와 오차가 적으면 사분위편차를 적용하였다. 각각의 이상치 제거 모형을 구한 다음 실시간 속도를 도로별 소통상태의 따른 기준 속도에 따라 원활, 지체, 정체로 구분하였으며, 본 연구의 분석구간의 소통 등급별 속도는 원활은 40㎞/h 이상, 지체는 40㎞/h 미만 20㎞/h 이상, 정체는 20㎞/h 미만이며, 이후 데이터 수집건수에 따라 빈도가 가장 많은 가중치를 적용하여 이상치를 제거한 다음, 통행속도를 산정하였다. 이를 토대로 이상치 제거 알고리즘을 제시하였으며, 알고리즘의 성능을 평가하기 위해 분석구간의 첨두일인 금요일에 GPS를 장착한 차량을 이용하여 모의주행을 시행하였으며, 그 결과 현장 조사시간대의 UTIS 수집건수는 총 167개였고, 관리도 모형은 이상치 제거가 없었으며(0%), 사분위편차의 경우 7개(4.1%)를 제거하였고, 개선된 알고리즘은 13개(7.7%)를 제거하였다. MAPE는 사분위편차가 58.8%, 관리도 모형이 60.5%, 개선된 알고리즘이 60.8%로 개선된 알고리즘이 기존 이상치 제거 모형보다 우수한 것으로 나타났다. 일부 이상치 제거된 시간대의 정확도가 높지 않은 경우도 발생하였으나, 다른 이상치 제거 방법과 차이는 크게 나지 않았으며, 대부분 이상치가 제거가 됨으로써 기존 이상치 제거 방법보다 정확도가 높아진 것을 확인할 수 있었다. 이는 실시간 5분 단위 데이터의 이상치 제거 시 개선된 알고리즘을 사용하여도 좋을 것으로 판단되며, 이상치가 제거된 데이터를 이용하여 지점검지기체계와의 퓨전을 통해 통행속도 및 통행시간을 산정하여 신뢰성 높은 정보를 제공할 수 있다.|There are UTIS (Urban Traffic Information System) and ITS (Intelligent Transportation System) for the improved urban traffic by providing qualified traffic information to users through collecting and analysing exact traffic information with connecting those devices which can build the traffic system offering information via media. In the case of UTIS, however, there are lots of researches to remove outlier in order to solve the problem of decreased reliability on traffic information resulting from various reasons including parked personal or business automobiles owing to their affairs or unexpected situation and duplicated data from abnormal wireless communication devices. But the outlier in those researches has been removed by applying the value of range equally from the quartile deviation and management graphic statistics model and this would effect the facts because the system provides information by collecting and processing the data every 5 minutes. In this respect it needs to identify the influence of each value of range on outlier filter by comparing and analysing them. Therefore this study has the purpose of providing proper methods accordant to collected data every 5 minutes and the range of outlier filter using quartile deviation and management graphic statistics model for outlier filter of real-time OBE data so that it can remove the outlier from the processed data every 5 minutes on individual automobile and increase the reliability of the information. For this purpose it selected a single intersection in Ansan-si and used OBE source data from September 01 to 30, 2011 collected by Hanyang University Traffic Information Center. The data was made by average speed per 5 minutes from the past and then the real-time average speed of 5 minutes and median were calculated. When the real-time and the past average speed had the low level of error on the base of the past average speed, management graphic statistics model was applied and when median and the past average speed had the low level of error, the quartile deviation was applied. After building the outlier filter model for each case, it divided real-time speed into smoothness, delay, and congestion accordant to the base travel speed. In this study smoothness means 40㎞/h or above, delay means 20㎞/h or above, and congestion means less than 20㎞/h. The travel speed was deduced by removing outlier after applying the most frequent weighted value from the collected data. On the base of the data above, outlier filter algorithm was provided. To estimate the function of the algorithm, there was driving simulation on the first Friday of the research using a car with GPS and total 167 was made for UTIS. Management graphic statistics model had no outlier filter (0%), the quartile deviation had 7 outlier filter (4.1%), and the improved algorithm had 13 outlier filter (7.7%). MAPE showed that the improved algorithm was superior to the precedent ones by making the result of 60.8% comparing the quartile deviation 58.8%, and management graphic statistics model 60.5%. There was somewhat low level of accuracy in a certain time zone but it was not high in comparing that of others and it was confirmed that the accuracy was more improved than the precedent methods by removing most outlier. The result of research suggests that the improved algorithm would be better for outlier filter of real-time 5 minutes data and it can offer information with higher reliability by calculating travel speed and time with fusion of position detection system using data with outlier filter.; There are UTIS (Urban Traffic Information System) and ITS (Intelligent Transportation System) for the improved urban traffic by providing qualified traffic information to users through collecting and analysing exact traffic information with connecting those devices which can build the traffic system offering information via media. In the case of UTIS, however, there are lots of researches to remove outlier in order to solve the problem of decreased reliability on traffic information resulting from various reasons including parked personal or business automobiles owing to their affairs or unexpected situation and duplicated data from abnormal wireless communication devices. But the outlier in those researches has been removed by applying the value of range equally from the quartile deviation and management graphic statistics model and this would effect the facts because the system provides information by collecting and processing the data every 5 minutes. In this respect it needs to identify the influence of each value of range on outlier filter by comparing and analysing them. Therefore this study has the purpose of providing proper methods accordant to collected data every 5 minutes and the range of outlier filter using quartile deviation and management graphic statistics model for outlier filter of real-time OBE data so that it can remove the outlier from the processed data every 5 minutes on individual automobile and increase the reliability of the information. For this purpose it selected a single intersection in Ansan-si and used OBE source data from September 01 to 30, 2011 collected by Hanyang University Traffic Information Center. The data was made by average speed per 5 minutes from the past and then the real-time average speed of 5 minutes and median were calculated. When the real-time and the past average speed had the low level of error on the base of the past average speed, management graphic statistics model was applied and when median and the past average speed had the low level of error, the quartile deviation was applied. After building the outlier filter model for each case, it divided real-time speed into smoothness, delay, and congestion accordant to the base travel speed. In this study smoothness means 40㎞/h or above, delay means 20㎞/h or above, and congestion means less than 20㎞/h. The travel speed was deduced by removing outlier after applying the most frequent weighted value from the collected data. On the base of the data above, outlier filter algorithm was provided. To estimate the function of the algorithm, there was driving simulation on the first Friday of the research using a car with GPS and total 167 was made for UTIS. Management graphic statistics model had no outlier filter (0%), the quartile deviation had 7 outlier filter (4.1%), and the improved algorithm had 13 outlier filter (7.7%). MAPE showed that the improved algorithm was superior to the precedent ones by making the result of 60.8% comparing the quartile deviation 58.8%, and management graphic statistics model 60.5%. There was somewhat low level of accuracy in a certain time zone but it was not high in comparing that of others and it was confirmed that the accuracy was more improved than the precedent methods by removing most outlier. The result of research suggests that the improved algorithm would be better for outlier filter of real-time 5 minutes data and it can offer information with higher reliability by calculating travel speed and time with fusion of position detection system using data with outlier filter.