Phase III: Meta Rules and Knowledge Class

Figure 15. Flow Chart of Intersection delay

4.4. Phase III: Meta Rules and Knowledge Class

The concept of Phase III, as shown in Figure 16, discusses about the interferences (attributes) of TTP and the building of knowledge class for later TTP expert system.

Also, the transformation of the results of traffic patterns (ID and STP patterns) from Phase II into travel time rules will be described. The construction of meta-rules, which can dynamic control the variables (αandβ) on the travel time rules of the historical and real-time travel time estimations, will be discussed at last.

Figure 16. Concept of Phase III

4.4.1. Interferences and Attributes of Travel Time Prediction

As shown in Figure 16, the vehicle starts at an origin position to reach its destination has many attributes for computing our TTP, and some of these attributes (facts) are regarded as interferences, which might delay the vehicles to reach the destination. Here, we list and briefly describe all attributes of TTP. First, Time and Location are not only the important index to present traffic information but also the

basic materials to formulate our traffic patterns and travel time rules. As we mentioned above, in this thesis, 1~48 time index to present 24 hours a day: 1 present 00:00 to 00:30, 2 present 00:30 to 01:00 and so on. The Location in this thesis is considered road sections in Taipei urban network as Figure 6, such as the sections of Zhong Xiao East Road. The Direction of LBS-based vehicles is also an attribute of TTP. Then, after considering time and location facts, the most important fact is Traffic Status, which has level 1-9 of average traffic speed. Then, we use the above facts to construct our STP knowledge.

In TTP, some geometry information is necessary material, such as Road Length and the Coordinates (X, Y) of intersection in GIS map. Because system can not compute the link travel time when it has the average speed in the target road but do not know the length of road. Thus the length of each section (Link) is needed in our target urban network. The coordinates (X, Y) of intersection are referred to GIS map and GPS position. These longitude and latitude of intersections are necessary for TTP system to precisely compute the ID patterns. The default patterns are used to handle missing or no related historical patterns (STP) in historical database. According to the Directorate General of Highways, we use some classified roads for making our default pattern. For example, the speed limit of freeway is 100KM/HR and second main line of urban network is 40KM/HR, etc. Then, TTP system can use this limit speed and the length of road to compute link travel time. Real-Time Events are the significant interferences of TTP. Here, we take incidents, road constructions and heavy raining events into account.

Types of Day, such as holiday and workday are also considered. Because different

traffic flows in the traffic network will have different traffic patterns, and we need to separate it for making our historical TTP more accurate. At last, Meta-rules and Knowledge Class of traffic patterns (STP, ID patterns) are also the portions of our TTP considerations, which will be discussed in following sections.

4.4.2. Generation of Travel Time Rules

After data mining process in Phase II, the three knowledge traffic patterns were found. Some of them are meaningful in TTP but some are not. In our goal of TTP expert system, we only use the STP knowledge and transform it into knowledge class [13].

Because, STP knowledge can be easily transformed into travel time rules for the TTP inference at run time by combining the link attributes in the road network table. Also, the

intersection delay patterns will be transformed into knowledge classes. And these knowledge classes can be used in the inference engine of TTP expert system at Phase IV.

Here are some transformation examples of STP and ID patterns as shown below.

„ STP to Travel Time Rules

STP- (Date, Time index, Holiday, Location, Dir. , Traffic level, Sup., Conf.) Here, we assume min. sup. =0.7, min. con. =0.8,

Ex: STP - (1, 16, 0, A1, 1, 4, 70%, 85%)

→ If Workday & 8am & A1 E. , then traffic level 4 …… (Travel time rule)

First, according to STP format in (3), time index, holiday, location, direction, and traffic level are chosen to transform the STP into travel time rules. Then, the “if…

then…” style is used to formulate our target travel time rules. The fact in RHS of travel time rules is only the traffic level, and the remaining facts are in the LHS of travel time rules. The thresholds of minimum support and minimum confidence are given by human expert for making our travel time rules more robust.

The transformation of ID patterns can refer to the format of ID patterns as (6). Every slot of ID format can be used to formulate travel time rules, where the fact in RHS of travel time rules is only the delay value and the other facts are in the LHS of travel time rules. The example is shown below:

„ ID Patterns to Travel Time Rules [TD/LTD/RTD]: (P, SOid, SIid, Tid, Davg) Ex: TD - (1, A1, B1, 1, 50)

4.4.3. Meta Rules Construction

The meta-rule is designed as a reaction mechanism to the current external traffic or non-traffic events in order to raise the precision of the real-time TTP in this section. In equation (1) and (2), control variablesαandβrepresent the weight of real-time and historical TTP respectively. Meta-rule, which is extracted from the expert, is the tuning mechanism for weighted combination of real-time and historical TTP. That is, meta-rules dynamically tune the value of weight control variables:αandβ. For example, if system receives a current external event, such as car accident on a link in the selected path, meta-rule mechanism will then reduce the weight of historical and raise the weight of real-time TTP. Because the effect of that car accident can reflect at that link immediately, so raising the ratio of real-time TTP can get higher precision. Here, the meta-rule is shown below and the style of meta-rules is as same as “if… then…”

format of travel time rules. In addition, the initial values of αandβ are given by human expert. In order to emphasize the real-time traffic status is more important in our real-time TTP system, the initial value ofαandβare set 0.7 and 0.3 by human expert, respectively.

[Meta-Rule]: If link is under construction, then α+=0.05 , β-=0.05

On the other hand, some meta-rules may raise the weight of historical TTP in several conditions: One, if there is no event happening or lacks real-time traffic information of LBS-based vehicles, Two, if the support and confidence values of the related traffic patterns (mining from the historical database) are higher than the thresholds. It means that there is a strong support that traffic status is possible to regress to the intents of related historical traffic patterns. Therefore, raising the weight of

historical TTP might get higher precision of TTP. The general format of this type meta-rule is like:

[Meta-Rule]: If link covers fewer probing vehicles, then α-=0.05 , β+=0.05