Beruflich Dokumente
Kultur Dokumente
Xuan-Yin Wang
State Key Laboratory of Fluid Power
Transmission and Control, Zhejiang
University, China
xywang@zju.edu.cn
1. Introduction
Abstract
635
Pm
Pm
0
favr f fmax
k
k
k
k
,
,
8
k
where, fset is the population maturity set point.
3.1. Reproduction
The reproduction number is computed by the
individual fitness. Supposedly, fi_p is the fitness of the
individual P, fi_sum is the sum of the population fitness, N
is the number of the individual in the population, Np is the
reproduction number of P in the next generation.
Pm
Pm
0
,
k
favr f fmax
636
3.2. Crossover
The operator is a neighborhood individual crossover,
which will work when Pr<Pc, where Pr is the complete
survival probability of the individual P, and Pc will be
computed by Form. 7. The operator is shown in Form. 1.
Figure 4. Single-joint control system diagram
3.3. Mutation
Supposedly, P distributes in [Pmin, Pmax] randomly,
when Pr<Pm,
(10)
where, Pm is computed by Form. 8, P is the new
individual after the mutation.
JL
JR
k
bL
bR
k k
11
3.4. Matching
When the population is mature, replace the individual
according a descending order of the individual fitness and
match the first pair of different individuals to do a
crossover operation, and then reproduce the individuals
for the second time to form the population again.
The main viewpoint of the MAGA is, when the
population is immature, let the individuals with low
fitness have high crossover probabilities to make the
population converge quickly, when the population is
mature, let the individuals with high fitness have high
crossover probabilities to expand the superiority genes, at
the same time, add the matching crossover between
individuals with the best fitness and the second best
fitness to do a neighborhood search, and then reproduce
the individuals to complete the population again.
w |
| d
12
where, u(t), the control signal; e(t), the system error; tu, the
system rising time.
wi(i=1,2,3,4) is the weight coefficient. In order to
suppress the system overshoot, a punishment coefficient
w4 is set, when e(t)0, w4=0, when e(t)<0, w4>>w3. In
the simulation, w1=0.999, w2=0.001, w4=100, w3=2.
13
637
Ki
Kd
0.0004 0.061
0.0004
0.0004
0.0004
0.0004
0.0004
0.0004
0.0004
0.0004
0.0004
0.0878
0.074
0.086
0.075
0.067
0.141
0.075
0.055
0.074
J
13.753
16.107
14.776
15.974
14.916
14.158
21.447
14.936
13.536
14.776
PAGA
Kp
33.504
35.951
27.701
22.228
30.550
32.031
35.951
28.065
33.548
35.951
Ki
0.0004
0.0004
0.0004
0.0004
0.0004
0.0004
0.0004
0.0004
0.0004
0.0004
5. Experiments
Kd
0.052
0.051
0.059
0.071
0.056
0.054
0.051
0.057
0.053
0.051
MAGA
Kp
23.759
26.390
34.369
31.396
29.353
30.876
27.603
33.3634
37.745
32.294
J
13.490
13.493
13.658
14.544
13.539
13.499
13.493
13.594
13.490
13.493
Ki
0.0002
0.0001
0.0001
0.0002
0.0001
0.0003
0.0003
0.0001
0.0003
0.0003
0.065
0.075
0.055
0.065
Kd
0.060
0.056
0.050
0.054
0.054
0.049
0.055
0.052
0.061
0.050
J
10.805
10.030
9.246
11.508
10.311
12.214
11.960
10.223
12.788
12.149
0.07
0.06
0.055
0.045
0.05
0.045
0.035
0.04
0.035
0.03
0.025
0.025
0.02
0.015
0.015
0.01
0.005
0.005
0
-0.005
-0.005
0
10
20
30
40
50
60
70
80
90
100
10
20
30
40
50
60
70
80
90
100
0.4
0.45
0.5
(b) PAGA
(a) SAGA
0.1
0.095
0.09
0.085
0.08
0.075
0.07
0.065
0.8
0.06
0.055
0.05
0.6
0.045
0.04
0.035
0.03
0.4
0.025
0.02
0.015
0.2
0.01
0.005
0
-0.005
10
20
30
40
50
60
70
80
(c) MAGA
90
100
0.05
0.1
0.15
0.2
0.25
0.3
0.35
638
6. Conclusions
Through the study on the past adaptive algorithm, a
matching crossover real-coded adaptive genetic algorithm
is proposed in this paper, which the crossover probability
changes according to not only the individual fitness but
also the population maturity, and a neighborhood search
method by adding the matching crossover to the algorithm
is also presented to decrease the search size of the realcoded genetic algorithm. The experiments on the PID
parameter optimization of a 6R series arc welding
manipulators demonstrate that this algorithm can generate
the superiority individuals more effectively and keep the
population diversity at a high level at the same time. To a
certain extent, this algorithm can solve the premature and
big search size problem and enhance the overall accuracy
of the real-coded genetic algorithm.
7. References
[1] Aidan ODwyer, PI and PID controller tuning rules:
an overview and personal perspective, ISSC 2006 Dublin
Institute of Technology, June 28-30.
639