Beruflich Dokumente
Kultur Dokumente
among the numerous responses estimated by the identified model, using a genetic algorithm.
The repetition of these operations leads to an adaptive growth optimization.
3.1 Plant growth responses and those identification results by way of neural network
Figure 3 shows daily growth changes in TLL/SD, amount of light intensity and three
operations of the nutrient concentration. It is found that the response of TLL/SD is markedly
affected by nutrient concentration. The data number N is equal to 22. Using this data, we
identified the TLL/SD as affected by both light intensity and nutrient concentration.
Figure 4 shows comparisons of the estimated response, obtained by the dynamic method,
and observed response of the TLL/SD. The system parameter number and the hidden neuron
2980
number for the neuro-computation were estimated to be 5 and 5, respectively. Hence the
neuron numbers of this neural network are 17 in the input layer, 5 in the hidden layer and 1
in the output layer. All estimated responses show a good agreement with respective observed
responses. Thus we could obtain successful model for the computation of the fitness (
TLL/SD ) to any 4-step values of the nutrient concentration.
3.2 Evolution characteristics by way of genetic algorithm and optimal control pelfoimance
The process of searching a optimal value is alike to evolution process. Figure 5 shows
evolution curves of the fitness ( TLL/SD ). Horizontal axis shows the iteration number for
genetic operations. The crossover rate P, and mutation rate P, used here are 0.8 and 0.8,
respectively. From the figure, the fitness dramatically increased with generation number and
then reached the maximum value at 9th-generation. In this case, more rapid increase in the
fitness are observed when both values of P, and P, take high values as shown in Fig.5. Thus
the use of genetic algorithm induced very quick search of the optimal value without falling
into local optima.
Figure 6 shows optimal control performance of the TLL/SD obtained from the calculation
of the dynamic model. The operation recommends to keep the nutrient concentration a little
high in the 1st-stage, markedly low in the 2nd-stage, a little high in the 3rd-stage and high
in the 4th-stage. From the regression method also, similar operation was obtained. The
application of it to the actual growth control resulted in a good plant growth.
4. CONCLUSION
The control technique combining genetic algorithm and neural network was applied to the
control for growth optimization of plant. The aim is to decide optimal 4-step operations of
the nutrient concentration of the hydroponic solution, which maximize the total leaf length
to stem diameter ( TLL/SD ). The use of neural network made it possible to identify such
complex system as the TLL/SD to nutrient concentration, and we obtained successful model
for computation of the TLL/SD. Furthermore the genetic algorithm enables us to quikely
search the optimal 4-step operations which maximize the TLL/SD among numerous responses
calculated from the model. The optimal value was obtained within 9-generation under high
crossover and mutation rates. This operation was available for the actual growth control.
Thus the hybrid control technique proposed here seems to be effective for solving the optimal
control problem of the plant to which mathematical approaches cannot apply.
REFERENCES
Chen, S., S.A. Billings and P.M.Grant ( 1990 ). Non-linear identification using neural
networks. International Journal of Control, 51(6), pp.1191-1214.
Genetic algorithms in search, optimization and machine
Goldberg, D ( 1989 ).
learning, Addison-Wesley.
Grefenstette, J.J. ( 1986 ). Optimization of control parameters for genetic algorithms.
IEEE Transactions on Systems, Man, and Cybernetics, 16(1), pp.122-128.
Morimoto, T., I. Cho and Y. Hashimoto ( 1991 ). Identification of hydroponics in an
advanced control system of the greenhouse. Proc. 9th IFACIIFORS Symposium
on Identification and System Parameter Estimation, 1, pp.610-615.
Morimoto, t., T. Torii and Y. Hashimoto (1993). Computer integrated system for plant
factory. Proc. 12th IFAC World Congress, ( in press ).
Van Winden, C.M.M. ( 1988 ). Soilless culture technique and its relation to the
greenhouse climate. Acta Horticulturae, 229, pp.125-132.
2981
optimization of plant
Time o f day
"'_cl-v
300
.
c
2
2
290
'
10
20
30
300 n
5 200A
c-
100 J
0-7
+ + I
"
c m
0.01,
10
Time o f day
15
20
30 0
I
z s
$"OR
alL
I ..-#
LE1+al
50
/--
2-
v)
v
40