Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
a r t i c l e i n f o a b s t r a c t
Keywords: This paper presents an improved self-adaptive particle swarm optimization algorithm (ISAPSO) to solve
Hydrothermal scheduling hydrothermal scheduling (HS) problem. To overcome the premature convergence of particle swarm opti-
Particle swarm optimization mization (PSO), the evolution direction of each particle is redirected dynamically by adjusting the two
Constraint handling sensitive parameters of PSO in the evolution process. Moreover, a new strategy is proposed to handle
Self-adaptive
the various constraints of HS problem in this paper. The results solved by this proposed strategy can
strictly satisfy the constraints of HS problem. Finally, the feasibility and effectiveness of proposed ISAPSO
algorithm is validated by a test system containing four hydro plants and an equivalent thermal plant. The
results demonstrate that the proposed ISAPSO can get a better solution in both robustness and accuracy
while compared with the other methods reported in this literature.
Ó 2011 Elsevier Ltd. All rights reserved.
0957-4174/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.eswa.2011.08.007
Y. Wang et al. / Expert Systems with Applications 39 (2012) 2288–2295 2289
function method, the proposed strategy did not need multiple run (5) Hydro plant discharge limits
for turning the penalty factors. And the constraints of HS problem Q hj;min 6 Q thj 6 Q hj;max j ¼ 1; 2; . . . ; Nh ; t ¼ 1; 2; . . . ; T ð7Þ
can be handled by proposed strategies without imposing any
restrictions. The proposed ISAPSO algorithm tested by a hydrother- where Qhj,min and Qhj,max are the minimum and the maximum
mal system contains four hydro plants and an equivalent thermal discharge of the j-th hydro plant.
plant. The feasibility and efficiency of proposed ISAPSO are proved (6) Reservoir storage volumes limits
by simulation results. It is found that ISAPSO can get a better solu- V hj;min 6 V thj 6 V hj;max j ¼ 1; 2; . . . ; Nh ; t ¼ 1; 2; . . . ; T ð8Þ
tion in feasible time, compared with the other optimization meth-
ods reported in the literature. where Vhj,min and Vhj,max are the minimum and the maximum
This paper is organized as follows: HS problem is formulated in volumes of the j-th hydro plant.
Section 2. In Section 3, the orginal PSO is described briefly. The pro- (7) Water dynamic balance
posed ISAPSO algorithm for HS problem is described in Section 4. Nj
X ts
The system simulation is given in Section 5. The conclusion is made V thj ¼ V hj
t1
þ Ithj Q thj þ Q hm mj ð9Þ
in Section 6. m¼1
X
Ns Nh
X In this section, the procedure of proposed ISAPSO for solving HS
P tsi þ P thj ¼ PtD ð4Þ
problem is described in details. The main stages of ISAPSO include
i¼1 j¼1
initialization, constraints handling and updating. The procedure of
where PtD
is the total load demand at interval t. Ns and Nh are applying ISAPSO to solve DED problem can be summarized as
the number of thermal and hydro plants. follows.
(3) Thermal plant power operating limits
4.1. Structure of individuals
Psi;min 6 Ptsi 6 Psi;max i ¼ 1; 2; . . . ; Ns ; t ¼ 1; 2; . . . ; T ð5Þ
For an individual P, which consists of N generating units and T
where Psi,min and Psi,max are the minimum and the maximum
intervals, the array of control variable vector is described as
power generation of the i-th thermal plant. 2 3
(4) Hydro plant power operating limits Q 1h1 Q 1h2 Q 1hNh P1s1 P1s2 P1sNs
6 7
6 Q2 Q 2h2 Q 2hNh P2s1 P2s2 P2sNs 7
6 h1 7
Phj;min 6 Pthj 6 Phj;max j ¼ 1; 2; . . . ; Nh ; t ¼ 1; 2; . . . ; T ð6Þ P¼6
6 .
7 ð13Þ
6 .. .. .. .. .. .. .. . 7
4 . . . . . . . 7
.
5
where Phi,min and Phi,max are the minimum and the maximum
Q Th1 Q Th2 . . . Q ThNh PTs1 PTs2 PTsNs
power generation of the j-th hydro plant.
2290 Y. Wang et al. / Expert Systems with Applications 39 (2012) 2288–2295
4.2. Constraints handling current interval with (7) satisfied, the largest volume at last interval
V t1 t1
hj;l arg is compared with Vhj,min. V hj;l arg is calculated by
In the operation of initialization or evolution, the discharges of Nj
X ts
hydro plants and the power generates of thermal plants may not V t1 t t
Q hm mj
hj;l arg ¼ V hj I hj þ Q hj;max ð17Þ
satisfy the various constrains of HS problem. The tradition con- m¼1
straints handling strategy for this case is penalty function method, t t
If V t1
hj;l arg is less than V hj;min ; Q hj ¼ Q hj;max , otherwise Q hj ¼ Q hj;min .
in which a penalty function is applied to convert a constrained
In the proposed strategy for inequality constraints, the back-
problem into an unconstraint one. Though the penalty function is
ward and forward operations adjust the discharge repeatedly until
convenient and easy-implemented, the procedure of multiple runs
the inequality constraints for hydro plants is satisfied or the max-
for the fine tuning of penalty factors would make a high computa-
imum trial number Maxcount is reached.
tion cost. Besides, the result of current solution may not always
fully satisfy the different constraints.
4.2.1.2. Handling the equality constraints considering the inequali-
In order to overcome the drawbacks of penalty function method
ties. Combining the Eqs. (9) and (10), we can get the equation
for this problem, a randomness adjustment strategy is proposed in
below
this paper. In the proposed strategy, the constraints of hydro plant
!
and thermal plant are handled respectively. For hydro plant, the X
T X Nj
T X
ts
discharge is firstly adjusted to satisfy the inequality constraints Q thj V hj;ini V hj;fin þ Q hm mj ¼0 ð18Þ
of this plant. Then, randomly adjust the discharge of different t¼1 t¼1 m¼1
interval in the feasible region until the equality constraint (10) is Recently, a simple method is proposed by Qin et al. (2010) to handle
satisfied. Considering that the operation of randomness adjust- this equality constraint is to add the difference of equality con-
ment is only taking effect in feasible region, thus the equality con- straint to one interval, which is described by
straints can be satisfied without breaking the inequality
Nj
constraints again. After the constraints of hydro plant are handled, X
T X
ts
X
T
the constraints of thermal plant can be handled similarly by the Q tshj ¼ V hj;ini V hj;fin þ Q hm mj Q thj ð19Þ
t¼1 m¼1 t¼1
strategy which is applied in the constraints of hydro plant. t – ts
Fig. 1. The flowchart of proposed constraints handling strategy for hydro plant.
2292 Y. Wang et al. / Expert Systems with Applications 39 (2012) 2288–2295
t
Where Violhj is the total violation of the j-th hydro plant, v iolhj is the 4.2.3. Total violation
volume violation of the j-th hydro plant at interval t, which is calcu- In proposed strategy, the violation is calculated while current
lated by individual cannot be adjusted to satisfy the various constraints of
8 t HS problem, the total violation including the hydro and thermal
>
> V V hj;max if V thj > V hj;max
< hj plants violation Viol is given by
t
v iolhj ¼ > 0 if V hj;min P V thj P V hj;max ð23Þ
>
:V Nh
X
Vt hj;min hj if V hj;min P V thj Viol ¼ Violhj þ k Viols ð30Þ
j¼0
8
< V hj;ini
> t¼0
where k is a factor. In this paper, k = 0.1.
V thj ¼ t1 t t P Nj
ts ð24Þ
>
: V hj þ Ihj Q hj þ Q hm mj others
m¼1 4.3. Rules of comparing two individual
Step 4: Find the thermal unit which are adjustable (For the x-th
In this paper, a new rule of updating the velocity is applied to
thermal unit, if Ptsx is adjusted deltat, and P tsx will not vio-
the procedure of Update and modification, the rule of each particle
late the constraints (5)). If there is no adjustable thermal
updating its velocity is described by
unit exist, turn Step 8, else turn Step 6.
8 gþ1 g
Step 5: Select one thermal unit randt from the adjustable unit g g g
> v i;t ¼ x v i;t þ c1 RandðÞ Pbest i;t xi;t þ c2 RandðÞ Gbesti xi;t
>
>
>
randomly. randt is the number of selected thermal unit. <
x ¼ ðxmax xmin Þ ebg þ xmin
Step 6: Adjust the randt-th generating unit with deltat, Ptsrandt ¼ >
>
> c1 ¼ ðd2 d1 Þ g=MaxGen þ d1
Ptsrandt deltat. >
:
Step 7: CountT = CountT + 1. c2 ¼ ðd1 d2 Þ g=MaxGen þ d2
Step 8: If CountT = n, turn Step 9, else turn Step 4. ð32Þ
t
Step 9: Calculate the violation of current interval v iols by where xmax and xmin are initial and final inertia weight factors, b is
!
X N XNh a shrink factor, d1 and d2 are constant factors. MaxGen is the maxi-
s t
v iols ¼ Psi PD Phj
t t t
ð28Þ mum generation number.
i¼1 j¼1 In (32), the two sensitive parameter of PSO c1 and c2 is changed
dynamically during the evolution procedure of proposed algo-
Step 10: If t is the last interval, turn Step11, otherwise, t = t + 1,
rithm. As the suggestion of (Kennedy & Eberhart, 1995), c1 + c2 =
turn Step2.
4. Generally a large c1 make the particles flying to the Pbest more
Step 11: Calculate the sum violation Viol and fitness value of this
probably. Similarly, and a large c2 make the particles flying to the
individual, where Viol is calculated by
Gbest more probably. In the proposed algorithm, d1 is initialed
X
T large than d2, which can make the proposed algorithm searching
Viols ¼ v iolts ð29Þ around Pbest more probably at the earlier stage of the evolution
t¼1
progress and around Gbest at the latter stage. For PSO, a large
Step12: the process of constraints handling is terminated. velocity makes the proposed algorithm explores globally, on the
Y. Wang et al. / Expert Systems with Applications 39 (2012) 2288–2295 2293
contrary, a small velocity will lead the algorithm searching in a lo- 4.6. Stopping criteria
cal area (Sun & Lu, 2010). So, to adequate the convergence of pro-
posed algorithm, x is initialed with a large value at the beginning The proposed algorithm is terminated while the maximum iter-
of run, and decreased rapidly in the process of evolution. ation number or the precision of solution is reached. Otherwise,
After updating and modifying the velocity and position of each continue evolution until the terminate condition is reached.
particle, the position of each particle may not satisfy the various
constrains of HS problem. In this case, the constraint handling 4.7. flow chart of proposed algorithm
strategy referred in Section 4.2 is applied to adjust the position
of individuals. The flow chart of proposed ISAPSO is showed in Fig. 2.
5. System simulations
start A test system which contains four cascaded hydro plants and an
equivalent thermal plant s is studied in this section to verify the
initialization
Table 1
Hydro plant data for the test system.
end
Hour 1 2 3 4 5 6 7 8
Demand (MW) 190 170 170 190 190 210 230 250
Hour 9 10 11 12 13 14 15 16
Demand (MW) 270 310 350 310 350 350 310 290
Hour 17 18 19 20 21 22 23 24
Demand (MW) 270 250 230 210 210 210 190 190
Table 4
The parameter of proposed algorithm for this test stem.
Table 5
The parameter of proposed algorithm for this test stem.
Table 6
The result of best fuel cost obtained by ISAPSO.
Interval(t) Water discharge (103 m3/h) Hydro generation (MW) Thermal generation (MW) Fuel cost Total
Qh1 Qh2 Qh3 Qh4 Ph1 Ph2 Ph3 Ph4 Ps f(Ps) RPh+Ps
1 5 6 21.07 13 23.73 16.76 42.91 45.62 60.99 3469.95 190
2 5 6 19.88 13 24.90 17.30 39.47 41.60 46.72 2558.65 170
3 5 6 18.7 13 26.03 17.84 37.39 44.14 44.60 2440.61 170
4 5 6 18.52 13 27.10 18.38 35.77 46.19 62.56 3582.39 190
5 5 6 17.85 13 28.13 18.90 33.98 47.83 61.17 3482.57 190
6 6.22 6 17.58 13 33.95 19.42 32.43 49.35 74.86 4550.66 210
7 8.28 6 17.41 13 42.24 19.92 31.25 50.63 85.95 5553.49 230
8 9.9 6.69 17.45 13 47.91 22.45 30.71 51.80 97.13 6688.51 250
9 11.21 7.83 17.44 13 51.79 25.92 30.56 52.89 108.85 8012.16 270
10 13.58 9.95 17.66 13 57.46 31.02 31.01 53.94 136.57 11690.75 310
11 15 11.93 18.39 15.01 59.35 34.23 32.42 59.34 164.66 16202.66 350
12 13.35 10.12 18.4 14.3 54.45 29.62 34.36 58.53 133.05 11181.21 310
13 15 11.91 18.84 17.21 56.71 32.23 36.39 64.92 159.76 15358.52 350
14 15 12.43 19.32 18.43 55.06 31.70 38.21 66.98 158.06 15071.85 350
15 13.59 11.26 19.41 18.2 50.77 28.35 40.14 66.78 123.96 9922.53 310
16 12.61 10.82 19.43 18.54 47.56 26.49 41.72 67.53 106.70 7758.96 290
17 11.4 10.27 19.2 18.92 43.96 24.56 42.66 68.25 90.57 6007.63 270
18 10.16 9.55 18.72 19.23 40.39 22.43 43.10 68.76 75.33 4590.29 250
19 8.37 8.87 17.99 19.5 35.22 20.55 42.99 69.05 62.19 3555.99 230
20 6.44 8.11 16.65 19.87 29.32 18.62 41.77 69.22 51.07 2814.68 210
21 6.81 8.58 15.52 20.97 31.51 19.61 40.27 69.61 48.99 2689.92 210
22 6.98 9.68 14.57 22.21 32.90 21.62 38.87 68.96 47.65 2611.55 210
23 6.1 6 13.97 23.19 30.47 12.70 38.07 66.94 41.83 2293.23 190
24 5 6 15.03 24.42 27.10 13.30 40.15 63.67 45.78 2505.86 190
Fig. 4. Convergence progress of PSO and ISAPSO for this test system. Fig. 6. Reservoir storage volumes for this test system.
Y. Wang et al. / Expert Systems with Applications 39 (2012) 2288–2295 2295
References