Conclusions

Global traffic network performance enhancement is one of the major tasks in ATMS, and it is the common hope of all the traffic network travelers as well as the network administrators.

However, this task is not only complex but also difficult due to several factors: network complexity in urban area, real-time sensor data unavailability, spatiotemporal traffic demand dynamics, spatiotemporal traffic bottleneck, and traffic demand conflict. These factors result in that traditional DTA-based approaches are hardly practical to real urban network.

In this dissertation, a novel approach based upon different philosophy of thinking is proposed, which builds the solution based on the concept of “continuously enhancing the global network performance” rather than the “optimal traffic assignment” concept in the traditional DTA-based approach. We try to enhance the global traffic network performance by four facets, including decreasing the traffic demand, discovering traffic patterns and identifying traffic bottleneck, resolving the traffic bottlenecks, and creating a collaborative traffic information generation and sharing framework for two-ways information exchanging between the travelers and the backend traffic information center. A series of works toward to the goal of enhancing the global network performance are discussed in this dissertation, including traffic information collection (Section 2.1), spatiotemporal traffic patterns and bottleneck discovering (Chapter 5 and Chapter 6), traffic information generation and sharing

100

framework (Chapter 4), multi-sources heterogeneous traffic information fusion (Section 4.3), traffic knowledge to inference rules transformation, travel time prediction system (Chapter 7) and ATMS decision support system (Chapter 8). These works are organized as a three-layered traffic knowledge framework, as illustrated in Figure 9-1, which includes data process layer, data mining layer, and application layer. As shown in the Figure, data collection, cleansing and traffic information generation are categorized in the data process layer, which collects raw data from the LBS-based applications, and the generated traffic information such as TIS, vehicle journey and TNS are stored in the TIDB. In the data mining layer, several hypothesis based spatiotemporal data mining models and heuristics are proposed for STPs mining and STB discovering to find out the traffic knowledge, which are transformed into rule classes and stored in the knowledge base. The traffic information, discovered traffic knowledge, domain expertise, and ontology constitute the hierarchical knowledge structure of the applications, as illustrated in Table 8-2. In the application layer, two practical applications are discussed: real-time travel time prediction system for travelers and ATMS decision support system for the network administrators. Knowledge based system technique is adopted in these two applications, where the discovered traffic knowledge combined with domain expertise constitutes the knowledge base, and real-time as well as historical traffic information, external traffic events data sources are regarded as facts in the inference engine.

101

Figure 9-1 Three-layer traffic knowledge framework

By the traffic information derived from LBS-based applications as discussed in Chapter 3, the vehicles in the LBS-applications can be regarded as traffic probing vehicles, which solve the cost, coverage, and real-time issues in traditional sensor-based or probing vehicle based surveillance system. In addition to traffic information collected and derived from the LBS-based applications, a wiki-like collaborative real-time traffic information collection and sharing framework based on the high penetration rates of location aware mobile devices is discussed in Chapter 4. By this collective traffic information generation scheme, more real-time traffic data will be collected cost-effectively and accurately, and the spatiotemporal coverage is better than the traditional traffic information collection scheme. Moreover, the framework integrates the generated traffic information with external traffic information sources into TIDB based on optimal weighting approach to minimize the data uncertainty, and

102

shares the traffic information with all the framework users by various data broadcasting channels.

The STP mining and STB discovering models discussed in Chapter 5 and Chapter 6 reveal the traffic bottleneck and traffic demand conflict issues in global traffic network performance enhancement task. Several spatiotemporal traffic patterns are defined and discovered from the TIDB based on the hypothesis based data mining techniques, and three heuristics are proposed to discover the STB in urban network by analyzing the discovered traffic patterns. On the other hand, CASR rules derived from the CPP can be a good explanation utility for describing the relationship between O-D traffic demand and congestions. Moreover, the discovered traffic patterns and O-D demand relationships present the congestion predictive capability, which provide decision support information for the ATMS administrator to take appropriate actions to solve the bottlenecks and thus enhance the global network performance.

In Chapter 7, a knowledge based real time travel time prediction (TTP) system is designed and implemented based upon the model proposed in this work, which contains real-time and historical travel time predictors and a dynamic weighted combination scheme.

The discovered spatiotemporal traffic patterns are transformed to the prediction rules in the historical travel time predictor, and real-time traffic information constitutes the real-time predictor and facts in the inference engine. Meta rules donated by traffic domain experts dynamically adjust the weight control variables for the linear combination of these two predictors by considering the effects and severity of the traffic events collected from the external traffic data sources. The experiments result show that real time predictor has better precision than historical predictor. By combining domain expert knowledge in meta-rules, weight combination predictor has better performance than other two predictors in both RME

103

and RMSE. In addition to the TTP system designed for the travelers, a knowledge based advanced traffic management system (ATMS) decision support system (DSS) targeted to improve the global network performance is discussed in Chapter 8, which provides valuable traffic assignment suggestions for the traffic network administrators. It utilizes the discovered traffic knowledge discussed in this work, such as traffic information database, STP/STB, predicted traffic status, and combines the domain expertise in order to enhance the global network performance and make traffic assignment suggestions. A hierarchical three-layered traffic assignment principles model combined with several traffic assignment principles is discussed to handle the different level of granularities of traffic patterns and traffic congestions. These traffic assignment principles donated by domain experts try to relieve traffic congestion and resolve traffic bottlenecks as well as traffic demand conflict.

Compromise concept is practiced in these principles to let the actions suggested by higher level principle have the major priority due to the higher level principles deal with the major traffic demands. The kernel of the ATMS DSS is the domain ontology which consists of two parts: meta-rules donated by the traffic domain experts and rules transformed from the spatiotemporal traffic patterns and bottleneck. The ATMS DSS is implemented as a knowledge-based system, where the rules are transformed from the ontology and facts come from the real-time traffic information in TIDB.

104