De Detec%ng and ec%ng and An Analyzing Urb Urban an Regions - - PowerPoint PPT Presentation
De Detec%ng and ec%ng and An Analyzing Urb Urban an Regions with High Imp mpact of Weather Ch Change on e on T Transpor ort Ye Ding, Yanhua Li, Ke Deng, Haoyu Tan, Mingxuan Yuan, Lionel M. Ni Presenta;on by Karan Somaiah Napanda,
Presenta;on by Karan Somaiah Napanda, Suchithra Balakrishnan, Zhaoning Su
manageme ment, data analy%c and service providing
*Urban compu;ng with taxis, MSRA *A Cloud-Based Knowledge Discovery System for Monitoring Fine-Grained Air Quality, MSRA
manageme ment, data analy%c and service pr providing viding
informa;on throughout a city
urban area
regions of different func;ons in a city
citywide petrol consump;on in real-;me…
to low visibility and high demand of
performance,
*Changes in Climate and Weather Relevant on US Transport, “The impact of climate change and weather on transport: An overview of empirical fin”
*Google Opera;ng System *IBM Smarter Traveler Traffic Predic;on So^ware
lacking of effec%ve traffic monitoring system in city-wide scale
throughout a city and factor analysis
numerous taxis driving on roads due to its availability, wide-coverage and low-cost.
record the informa;on including loca;on, speed, occupancy status, and orienta;on of the taxis
Voronoi diagram
*The Voronoi diagrams par;;ons in Shanghai. The under layer represents the road networks.
Equal-sized rectangles Ø some cells are highly dense and in others are highly sparse Ø seeds are the road intersec;ons Ø every cell include at least one road intersec;on and a number of roads connected to this intersec;on Ø if several road intersec;ons are very close to each other, for example within 50 meters, they are grouped together as a complex intersec;on
informa%on? In the area of residence community, everyone have their own car so they don’t need the taxi
in each Voronoi cell can reflect the average speed? However maybe some cells have a lower average speed of taxis is just because people are likely get on
how to disclose the key factors behind the weather-traffic index
density of roads number of road intersec;ons number of POIs(points of interest) traffic volume average age of the household density of buildings and more in the surrounding regions
To develope weather-traffic index (WTI) system
The first is to set up a weather-traffic index throughout a city, which indicates the impact of weather to traffic from light to heavy. The second is to reveal the key factors behind the weather-traffic index throughout the city and their rela;ve weights. Previous works mainly focus on the analysis of road segments; on the contrary, this paper is the first study on local traffic-weather sensi;vity throughout a city and the inves;ga;on to reveal the key factors behind the sensi;vity
VORONOI DIAGRAM Par;;oning of a plane into cells/ regions based on the distance between seeds Shape and size of each cells is different from each other SEEDS- road intersec;ons ROAD- INTERSECTION- ORIENTED Par;;oning Proper;es
traffic
i. Road Networks - G(V, E) E- set of road segments V- set of road intersec;ons E- type, length, speed limit, two end points V- loca;on (la;tude and longitude), type i. Road Networks ii. Traffic Data
ii. Traffic Data Traffic Parameters of interest are extracted
Time Mean Speed/ Average Speed- Traffic Parameter Arithme;c mean of individual spot speeds that are recorded over a selected ;me period Extracted from taxi trajectories in each cell Split into 7 classes
The average speed of one road segment is subject to the traffic parameter of that road segment only, not comparable with other road segments
Four Categories a) Points of Interest b) Structure c) Density d) Community
Input – Weather data and Traffic data Indicates the impact of weather to traffic in different cells Given a cell g, its value in the weather traffic index is the correla;on between traffic and weather, denoted as ρ(g). ρ(g) takes a value from the discrete range [1, 2, 3, 4, 5]
Correla;on between traffic speed Ft and weather Fw. Classifier is trained with Ft and Fw Input- Weather as a feature vector Output- one of the seven speed classes Inference accuracy Correla;on between Ft and Fw in that cell Inference accuracy Correla;on between Ft and Fw in that cell Weakness of this method, Does not consider other reasons that affect traffic
Granger Causality test whether one ;me series is useful in forecas;ng another A ;me series X is said to Granger-cause Y if it can be shown that those X values provide sta;s;cally significant informa;on about future values of Y. A variable X that evolves over ;me Granger-cause another evolving variable Y if predic;ons of the value of Y based on its own past values and on the past values of X are beuer than predic;ons of Y based only on its own past values.
for different ;me
each cell to indicate the extent to which traffic is impacted by weather
predic;on accuracy improvement and then divided into k- equal sized subsets
and weather from weak to strong
Support Vector Machine (SVM) Logis;c Regression (LOGIT) Perceptron
to verify the output of SVM
AssumpRon – weather traffic index of all cells have been certainly assigned This method iden;fies the key factors and their weights contribu;ng to the weather- traffic indices of cells. Discloses what regional features make the traffic in some cells vulnerable to inclement weather Steps:
located (or adjacent) cells
input, it indicates that such set of regional features are the key features
Marginal DistribuRon Probability distribu;on of the regions contained in a similarity subset Probability of one region being the index of i given one of its adjacent regions with index j, if two regions have a certain similarity Cosine Similarity is used to describe the similarity muv between two regions gu and gv Similarity ranges of k- groups b0 is minimum similarity Bk is maximum similarity
Pairs of cells in the group of [bi-1, bi] are in marginal distribu;on matrix Bi Rows of Bi- weather traffic indices of gu Columns of Bi- weather traffic indices of gv pij- probability that the weather-traffic index of gv is ρj when the weather- traffic index of gu is ρi
Index- Index Inference From marginal distribu;on, weather traffic index of a par;cular cell is inferred from adjacent cell using naïve Bayes classifier. Given a cell gu, the marginal distribu;on allows naïve Bayes classifier to infer which value the weather- traffic index of gu is most likely to be, based on the weather- traffic indices of its adjacent cells ρ(g1), ρ(g2),…
Consider a regional feature with nontrivial impact to weather-traffic index,
r) weight of the regional feature Fi r by compu;ng the
difference of the inference accuracy with and without Fi
r
§ Na;onal Expressway § City Expressway § Regular Highway § Large Avenue § Primary Way § Secondary Way § Regular Road Weather Report Data
Underground (wunderground.com)
the same ;me frame as taxi trajectory.
Major Roads Minor Roads
Interest) data.
informa;on about loca;on, price and residen;al communi;es.
merchant informa;on and other aurac;ons around the locality
What is the WTI
Robustness of the model:
used to test the robustness of WTI (with/without weather).
indicates low impact.
The above diagram shows most
accuracy of 0.5 for SVM. The above diagram gives changes in the predic;on accuracy which is the least for
the base model and Logis;c Regression and Perceptron are used to test. The changes in predic;on accuracy are close to mean for SVM and Logis;c Regression and it’s a regular distribu;on where as for Perceptron is greater.
that it has significant impact on traffic.
rain both have significant impact on traffic.
rain affec;ng traffic than weather affec;ng traffic.
rainy days.
gap between different likelihoods.
Regional House Age is an important factor