Page 126 - Special Topic Session (STS) - Volume 3

STS520 Xin Zheng et al.
the ARIMA model is used to predict the population size at the next time
period with the historical population size at p time periods.
(2) Using the spatial sequence correlation
Generally, the spatial correlation is given in space according to the
distance. If the Euclidean distance is used as the measure of distance, the
distance between any two places can be defined as follows:

By taking 1 ⁄ ( − ) as a weight, the population size of the central grid

can be estimated by the weighted mean value of population size of nearby
grids.
(3) Using the AI deep learning method
The deep residual ResNet method is a classical AI deep learning model.
Its basic principle is that the increase in the number of layers will improve the
learning effect of network in the neural network model, but at the same time
this may result in the increase of errors. In the deep residual ResNet method,
the residuals are learned instead of the mapping relation, so this problem is
well improved.

3. Result
(1) By taking Tiantongyuan area in Beijing City as an example, the population
stock recorded at 14:00 on the first Monday of each month is used as the
sample data, and then the related structure in time is verified by time series.
First, the data series will be subject to the white noise test. If the p value is
smaller than 0.05, it will be judged as a non-white noise series. Second, the
stationarity test is carried out to observe the changes in autocorrelation
coefficient and partial autocorrelation coefficient. The autocorrelation
coefficient does not attenuate rapidly, so we have reasons to believe that this
time series is a non-stationary series. To further verify our speculation, the
unit root test is introduced to judge the stationarity of the series. The p value
of 0.9585 is larger than the significance level of 0.05, so this series is judged
as a non-stationary series. Through logarithmic transformation and first-order
difference transformation, the new series will be subject to the unit root test.
The result shows that the p value is smaller than 0.05, so it is judged that the
series is a stationary series after processing. The model is automatically
determined by means of AIC and BIC statistics, and the ARIMA (0, 1, 1) model
is used for the new series. The first-order MA coefficient is significantly lower
than 0.05. The sample data is fit, and the result is shown in Figure 3. The blue
color represents the original value, and the red color represents the predicted
value. The autocorrelation coefficient and white noise test are adopted for
residuals, and the p value of autocorrelation test over residuals is larger than

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