Article Preview
Top1. Introduction
In recent years, with the rapid development of the Internet of Things (IoT) and mobile computing technologies, people have accelerated the time to step into big data. Many applications quickly generate a wide variety of data, and real-time and efficient processing of these data can yield greater value. The real-time processing of data streams is a key to big data research. Velocity and diversity are just the description of this situation in the characteristics of big data.
As the core of IoT middleware and the key technology of big data stream processing, complex event processing (CEP) is receiving extensive attention. Complex event processing is the process of interpreting and combining the original events in the data stream, higher-level complex events (also called compound events) is identified (Luckham D., 2011). Complex event processing technology has been widely used in many fields, such as wireless sensor network-based environmental monitoring and continuous analysis of stock movements. Recent large-scale streaming computing technologies have also been rapidly developed, such as Apache's Storm, Spark Streaming and Flink projects. These projects emphasize the efficient processing of large-scale distributed data streams, but do not directly support complex operations. Combining these technologies with complex event processing can better support the efficient and intelligent processing of distributed data streams.
The development of Internet of Things and big data streaming technology has brought new opportunities for the development of intelligent transportation systems (ITS). With the help of the Internet of Things technology, all-round perception and data transmission of traffic information can be carried out. Opportunity big data stream processing technology can process the large amount of the generated data in real time and intelligently. Traffic flow prediction is used as a basis for many traffic intelligence decisions, it occupies an important position in the intelligent traffic system.
People have conducted a lot of research on traffic flow prediction problems and have developed a variety of prediction models and methods. In recent years, there have been more successful methods based on support vector machines (Hong W C, et al., 2011;Yang Y N and Lu H P, 2010), methods based on deep learning (Huang W H,et al., 2014;Lv Y S, et al., 2015), methods based on Bayesian networks (BN) (Sun S L, et al., 2006;Castilloa E, et al., 2008;Zhu S L, et al., 2013) and hybrid methods (Wang J, et al., 2014) and so on. Based on Bayesian statistics technology, Bayesian network can well integrate domain knowledge and data, support the processing and causal analysis of incomplete data, and it can better solve the problem of overfitting. These characteristics make the Bayesian network occupy an important position in the prediction model and have been widely used in traffic flow prediction.
The current traffic flow prediction methods still have the following problems:
- 1.
When the environment changes, a fixed model cannot have good prediction ability in different environments (Wang J, et al., 2014). The method based on the hybrid model can overcome this problem to a certain extent and avoid the poor prediction results when the environment changes. However, the model combination method cannot guarantee the best results in each environment.
- 2.
The existing methods basically train models from a large amount of historical data, and then the models are used for real-time forecasting, but it seldom consider that the models may need to change correspondingly with the production of new data.
- 3.
With the development of the Internet of Things and mobile computing, people can obtain a variety of environmental information in real time, such as road conditions and weather. Existing methods of traffic prediction are often based on limited information and do not use a variety of real-time environmental information.