Research Papers

Study on Framework of Traffic Operational Safety Assurance Program and Real-time Crash Risk Prediction Model of Shenzhen-Zhongshan Channel

Filename 5C_3_Chen_Paper.pdf
Filesize 1 MB
Version 1
Date added June 28, 2017
Downloaded 1 time/fois
Category 2017 CARSP XXVII Toronto
Tags Research and Evaluation, Session 5C
Author/Auteur Yongsheng Chen
Stream/Volet Research and Evaluation

Slidedeck Presentation

5C_3_Chen

Abstract

"Shenzhen-Zhongshan Channel (Shen-Zhong Channel) is designed as a 24-km long channel across the sea outfall of Pearl River, Guangdong Province, China, combined with 2 man-made islands, a tunnel, and a bridge, which will connect directly two major cities: Shenzhen (population: 11.4 millions) and Zhongshan (population: 3.1 million). It is designed as the 8-lane freeway with design speed as 100 km/h, projected to be open by 2024. This mega channel will be encountered with significant traffic operational safety concerns due to extremely high traffic volume (over 90 thousands pcu/day), high portion (about 50%) of trucks, and extraordinarily high exposure of dangerous goods and over-sized vehicles. To tackle the above-mentioned concerns, a research project was funded by the future owner of the channel, which is dedicated to research and develop an operational safety assurance program with variable lane management, smart signal and speed control in order to minimize the crash and congestion risks. The foundation of this program is a real-time crash risk prediction model with traffic operational features, including volume, speed and density as inputs. Then the management and control modules of this program will be developed with reduction of risks as controlling objective. Since Shen-Zhong Channel itself is still at preliminary design stage, a freeway with similar traffic operational characters was selected as pilot road and its historical collision and traffic flow data were collected to develop the risk prediction model.

At first, the risk was defined as a variable with values between 0.0 and 1.0. At a moment a real collision occurred, risk value was quantified as 1.0. Prior to this moment, risk values were linearly reduced until down to 0.0. Then, A series of data mining and artificial intelligent algorithms, including Decision Tree, Logistic, Support Vector Machine (SVM), and Adaptive Neuro-Fuzzy Inference System (ANFIS), were attempted to develop models predicting risk values from traffic flow features. Finally, a specific 2-step mechanism was recommended: step 1, Decision Tree method was applied to obtain the major contributors to risk from a variety of traffic operational characters, and step 2, these major contributors were input into ANFIS to predict risks. After the risks were predicted, a threshold triggering alarm was set up as 0.6 based on historical observation. Once the risk was predicted above 0.6, there would be a “high risk” alarm. By this standard, 20 historical collisions were tested and among them 16 triggered the alarm. That is to say, this prediction model gained about 65% accuracy rate. There are still some aspects need to be addressed in future. The fault alarm rate needs to be checked as well. And also, the algorithm must be trained and calibrated by real Shen-Zhong Channel traffic data before it can really be applied into operation. In order to mitigate traffic operational safety concerns for Shen-Zhong Channel, a safety assurance program with variable lane management, smart signal and speed control was designed. The core algorithm of this program, real-time crash risk prediction model, was developed based on observed crashes and traffic operation data. "

Yongsheng Chen