129
Zhang
et al
.,
Improved online sequential extreme learning machine for simulation of daily reference evapotranspiration
Tecnología y Ciencias del Agua
, vol. VIII, núm. 2, marzo-abril de 2017, pp. 127-140
ISSN 2007-2422
•
matrix is adjusted with reference to the optimal
solution and the regularization factor at the
same time, which is motivated by the online
learning method. In summary, an improved
online sequential extreme learning machine
(IOS-ELM) is designed, and the new algorithm
can produce good generalization performance
in a model of daily
ET
0 in an irrigation system.
Materials and case study
The study was conducted in Yulin (38.27°
N, 109.78° E), Ankang (32.72° N, 109.03° E),
Hanzhong (33.07°N, 107.7°E) and Xi’an (34.3° N,
108.9° E) in Shaanxi province in China, shown
in Fig. 1. The area has a hot and dry climate for
the greater part of the year.
Daily meteorological data used for this
study was from the years 1971–2014. The fol-
lowing observed eight meteorological variables
with daily temporal resolution were used:
wind speed at 10m above the ground (
Uh
),
mean temperature (
T
), mean relative humidity
(
RH
), minimum temperature (
T
min), maximum
temperature (
T
max), actual Sunshine duration
(
n
), latitude (φ) and elevation (
Z
), which were
downloaded from China meteorological data
sharing service system
/
home.do). Data from the first 29 years (1971-
1999) was used to train the models. Data from
the next ten years (2000-2009) was used for the
test. The data from the remaining years was
used for validation. It must be noted, however,
that missing data was replaced by the average
of the data from the day before and the day
before. The regional climate characteristics are
given in table 1.
Figure 1. The location of the cities.
Methodology
Calculation models of reference crop
evapotranspiration
The study focused on the comparison of the pro-
posed IOS-ELM model with the ELM, LSSVM,
Table 1. Means of main variables.
Uh
T
RH T
min
T
max
n
φ
Z
Yulin
2.14
9.2
53.5
3.1
16.2
7.32
0.67
1157
An kang
1.35
15.9
74.1
12.1
21.4
4.58
0.57
290.8
Han zhong
1.15
15.1
78.5
11.5
19.8
4.03
0.58
509.5
Xi’an
1.60
14.6
64.3
9.7
19.6
4.54
0.60
397.5
