Page 131 - Special Topic Session (STS) - Volume 3

STS520 Xin Zheng et al.

Figure 7. Errors of Prediction Based on the Deep Residual Model.

Figure 8. Distribution of Errors.

4. Discussion and Conclusion
As the change of population size is periodic in time, if it is observed on a
daily basis, the population fluctuates from morning to night every day; if it is
observed on a weekly basis, there are also significant differences on working
days and at weekends; if it is observed on a yearly basis, the climate of four
seasons and the holidays exert an obvious impact on people flow. Therefore,
when the time correlation is used to predict the population size, to acquire a
better prediction effect, either short-term prediction or time series with fixed
intervals should be used to eliminate the impact brought by periodic factors.
If the deep residual model is used, part of the areas that show relatively
large errors are the areas where the people flow is large, and the other part
of the areas are the areas where the people flow is small, especially the urban
border areas. It can be seen that this model still needs to be improved for the
learning training and prediction of extreme value.

120 | I S I W S C 2 0 1 9