This is a paper recently published by our research group in the journal Communications in Transportation Research. This journal should be a pretty good journal. Next, let’s introduce our work.
1. Article overview
The main purpose of the article is to analyze how the travel behavior of urban residents will change due to the impact of the new crown pneumonia epidemic, and what factors are related to this change. The article is mainly based on the analysis of license plate recognition data and POI data. First, the development of the epidemic is divided into three stages, and then quantitatively analyze the changes in the travel behavior of travelers in the three stages. Based on the clustering algorithm, the behavior adjustment model is obtained, and on this basis, the impact of the travel behavior of travelers on the behavior adjustment is analyzed.
2. Travel mode analysis and adjustment mode analysis
Based on the license plate recognition data, the spatio-temporal travel chain of the traveler in each epidemic development stage is extracted, and then the frequent travel pattern set of the traveler is extracted by using the Prefix-Span algorithm. Since the frequent travel mode is often not one, because the frequent travel mode can be binomial sets, triitems, etc., and are therefore frequent travel pattern sets. Of course, there are many data processing processes in this process, including: 1. An item is a [x, y, t]. How to measure whether a space-time point is the same point as another space-time point requires some processing; 2. Due to The sample size is huge, and the result cannot be obtained within an acceptable time by using the Prefix-Span algorithm directly. Therefore, compression is required. We propose a compression algorithm. The whole process is shown in the figure below:
After obtaining the frequent travel patterns of each traveler at each stage, it is necessary to propose a similarity measurement method for frequent travel patterns in different stages and the regularity of travel behavior within a stage. We use the following formula:
At this point, the quantitative measurement of the frequent travel patterns of each traveler at each stage can be obtained, and then the behavior adjustment model can be obtained by using the clustering algorithm.
3. Analysis of influencing factors of behavior adjustment mode
This part is mainly to analyze how the previous travel behavior of the traveler affects the behavior adjustment mode. The idea is very clear, that is, the second part of the behavior adjustment mode label is used as the dependent variable, and some pre-epidemic behaviors are extracted as independent variables, and then randomized The forest algorithm analyzes the connection between the two. However, some intermediate processing has also been done, including: 1. RFECV for feature screening; 2. Borderline-SMOTE to solve the problem of sample imbalance.
After constructing the random forest model, Partial dependence plots are also used to analyze how each feature affects the behavior adjustment of travelers. The result is not displayed
4. Other
The paper also analyzes the details of various behavior adjustment modes, such as the change of the first and last travel time, the change of travel intensity, etc., of the people in various behavior adjustment modes. For details, please refer to the article.
references
Yao, W., Yu, J., Yang, Y., Chen, N., Jin, S., Hu, Y., Bai, C., Understanding travel behavior adjustment under COVID-19, Communications in Transportation Research, https://doi.org/10.1016/j.commtr.2022.100068.
Article link: https://www.sciencedirect.com/science/article/pii/S277242472200018X?via%3Dihub
The code is open source: https://github.com/RobinYaoWenbin/Understanding-travel-behavior-adjustment-under-COVID-19