International Core Journal of Engineering 2020-26 | Page 112

TABLE III. P REDICTIVE AND R EAL AQ GRADE ON A PRIL 1 ST
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
Num. error
Accuracy (%)
Beijing Beijing
Olympic
Centre Changsha
0/0
0/0
0/0
0/0
0/0
0/0
0/1
1/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/1
0/1
1/1
1/1
1/1
4
83.33 0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/0
0/1
1/1
1/1
1/1
1/1
1
95.83 1/1
1/1
1/1
1/1
1/1
1/1
1/1
1/1
1/1
1/1
1/1
1/1
1/0
1/0
0/0
0/0
0/1
1/1
1/1
1/1
1/1
1/1
1/1
1/1
3
87.5
Changsha
Southern
Train
Station
1/1
1/1
1/1
1/1
1/1
1/1
1/3
1/1
1/1
1/1
1/1
1/1
1/1
1/1
1/1
1/0
0/0
0/1
0/1
1/1
1/1
1/1
1/1
1/1
4
83.33
At the same time, by analyzing the causes of the errors, we
can find:
x The forecasting models respond slowly to sudden
changes in air quality grades. This is mainly reflected
in the comparison of the predicted results at 6 o'clock
in the morning at Changsha Southern Train Station.
Fig. 2. accuracy curves of Beijing Olympic Center and Changsha Southern
Train Station
x Forecasting models tend to produce hysteresis during
periods when air quality grades are prone to change.
This shows that the forecasting model needs to learn
incrementally.
C. ForecastingResults
In this section, the performance of the initial forecasting
models is tested on the predicting datasets. The initial
forecasting models first make predictions on a instances
without the real air quality grade. Then, the real air quality
grade is used to test the predictive accuracy of the ensemble
algorithms. Table 3 shows the predicting air quality grade and
the real grade on April 1st. And the comparison results in the
table is presented using the predictive/real format. For
example, the value 0/1 means that the predictive air quality
grade is 0 and the real grade is 1. If the predictive value is
different from the real value, the cell will be filled with yellow.
x Although the model is not sensitive enough to predict
the sudden change of air quality grade, from the 24-
hour continuous forecast, the prediction accuracy of
the method is more than 83%. It shows that this method
still has great application value in real life.
As it shown in Table III, the predictive accuracy is best at
the Beijing Olympic center. And most of the wrong prediction
happens from 12:00 to 20:00. In general, the inversion layer is
prone to occur at night, in the morning, and in the evening.
During these times, various pollutants in the air are not easily
diffused. Therefore, the air quality grade is likely to rise during
these periods. After the sun came out, the ground temperature
rise rapidly and the inversion layer begin to dissipate.
Therefore, after 10 am, the air quality will generally turn better.
Theoretical analysis is also confirmed in the results. First, air
quality grades tend to rise between 7 and 8 in the morning.
Similarly, during the period from 16:00 to 8:00, the air quality
grade tends to rise or fluctuate at each monitoring station in
each city. In the above table, it can be found that after 12
o'clock, the average air quality grade of the city of Changsha
has dropped from 1 to 0, and it did not rise until 16 pm.
Fig. 3. Forecasting results of LB prediction model of Beijing
Next, the results on the whole prediction dataset are
presented (From 22:00 on March 27, 2019 to 14:00 on April
90