.
There is another way of processing the data to improve
the generalization ability of the network, which is called the
normalization processing, as shown in Equation (12).
max( ( )) min( ( ))
( ) ( ) min( ( ))
d t d t
X t d t d t
−
= − (12)
Where d(t) is the data before normalization, X(t) is the data
after normalization.
Similarly, the output of the network has to experience a
counter-normalization processing, shown as in Equation (13).
Y (t) = o(t) × (max(d (t)) − min(d(t))) + min(d(t)) (13)
Where o (t) is the output value of the network, Y (t) is the
result after counter-normalization.
The purpose of Equation (12) and (13) is to restrict the
range of the input value to [0, 1], while in this network,
normalization processing can not reflect the real data directly
and the method of dividing the solar radiation by a constant is
used instead of the normalization processing.
E. Method Chossing
As the BP network uses the Error Back Propagation
which is a kind of unconstrained nonlinear optimum
calculation process, the calculation is rather complex and
time consuming and the optimum result is difficult to acquire
when the network is also complex. Although now there are
many improved BP algorithms, for example, the genetic
algorithm and the simulated annealing algorithm.
Theoretically, these algorithms can reach the global
minimum point by changing several parameters but it is rather
difficult in real applications. The traingdm, traingda,
traingdx, trainrp and trainlm are all improved BP algorithm
by adding the momentum term. In this proposed network, all
these methods are used to train the network and decide the
best one by considering the training speed and the errors of it.
The influence of the learning method to the neutral
network is far less than that of the training method; the
learndm is used in this proposed network.
IV. RESULTS AND ANALYSIS
A. Solar Radiation Data of Guangzhou
Guangzhou locates in the south of China and it is also a
massive manufacturing center. In recent years, many
photovoltaic systems have been built in Guangzhou,
including some wind/photovoltaic hybrid systems, to help
save the traditional fuels and better protect the environment.
Solar radiation data of Guangzhou is a very important
parameter of designing the photovoltaic systems and the total
energy generated by PV panels can be calculated before the
installations [13, 14]. The solar radiation data of each Chinese
city can be found available online at the “China
Meteorological Data Sharing Service System”. The yearly
total solar radiation data of Guangzhou from the year
1961-2003 are shown online. (Solar radiation data from the
year 2004-2007 have not been opened to the public).
The data from 1961 to 2000 are used to train the network
and the data from 2001 to 2003 are used to test the output of
219
the network. The training precision is set to be within ±5%
permit. since the data range is limited to [0.2, 0.8], each solar
radiation data must be divided by 10000.
B. Tanning Process and its Results
The training methods in MATLAB contain traingdm,
traingda, traingdx, trainrp and trainlm, etc. The network has
3 data for inspection, which is 0.4235, 4220 and 0.4607,
respectively. The epochs of training are set to be 1000 and the
time is set to be
本论文由英语论文网提供整理,提供论文代写,英语论文代写,代写论文,代写英语论文,代写留学生论文,代写英文论文,留学生论文代写相关核心关键词搜索。