Information Technology Reference
In-Depth Information
Algorithm 1.
New cells generation
1:
for all
newcells
do
2:
sum ←
0
for
i ←
1
,N
intervals
do
Random initialization
3:
4:
cell
(
i
)
←
random()
5:
sum
=
sum
+
cell
(
i
)
6:
end for
7:
for
i
←
1
,N
intervals
do
Normalization and interization
8:
cell
(
i
)
←
INT(
N
free
× cell
(
i
)
/sum
)
9:
end for
10:
end for
Algorithm 2.
Mutation
1:
for all
clones
do
2:
for
i ←
1
,N
intervals
do
3:
mutaz
(
i
)
←
random()
4:
if
0
≤ mutaz
(
i
)
≤
1
/
3
then
mutaz
(
i
)
←−
1
if
1
/
3
≤ mutaz
(
i
)
≤
2
/
3
then
mutaz
(
i
)
←
1
5:
if
2
/
3
≤ mutaz
(
i
)
≤
1
then
mutaz
(
i
)
←
0
6:
7:
end for
8:
for
i ←
1
,N
intervals
do
9:
clone
(
i
)=
parent
(
i
)+
mutaz
(
i
)
− mutaz
(
i −
1)
Feasible mutation
10:
if
clone
(
i
)
≤
xlow
(
i
)
then
Fix mutation to the lower bound
11:
clone
(
i
)
← xlow
(
i
)
12:
mutaz
(
i
)
←
0
13:
end if
14:
if
clone
(
i
)
≥ xup
(
i
)
then
Fix mutation to the upper bound
15:
clone
(
i
)
← xup
(
i
)
16:
mutaz
(
i
)
←
0
17:
end if
18:
end for
19:
end for
5
Proof of Principle Test Case
MILP and AIS-LP are tested on a simple but effective energy management prob-
lem. The structure of the CHP node is the one of Fig. 1; the operational data
of the devices are reported in Table 2. The thermal storage unit is considered to
have a maximum capacity of 300 kWh. Energy price profiles are shown in Fig. 4.
Several scheduling instances are solved with a quarter of hour time sampling
(
Δt
=0
.
25 hours), thus a one day scheduling period has
N
intervals
= 96, two
days scheduling
N
intervals
= 192 etc. Results are compared in terms of conver-
gence time and number of objective function calls. It must be remarked that a
comparison in terms of the mere number of objective function calls can be mis-
leading because the linear problem solved by MILP and AIS-LP are different.
