BATCH REACTOR OPTIMAL CONTROL USING GENETIC ALGORITHM

1
BATCH REACTOR OPTIMAL CONTROL USING GENETIC
ALGORITHM Dr. V. Lavric, Drd. O. Raducan and
Prof. A. Woinaroschy Department of Chemical
Engineering, University Politehnica of Bucharest,
RO-011061 Polizu 1-7, Bucharest, Romania
Problem Formulation
Genetic Algorithms Features

• Genetic operators
• Selection choose the fittest parents to be
  used to produce better offsprings
• Crossover recombine genes of randomly selected
  pairs of individuals with certain probability
• Mutation randomly change genes in the
  chromosomes with a very small probability (this
  keeps the population diverse and prevents from
  premature convergence).

• Main Concepts
• inspired from the process
• of natural selection of biological organisms
• representation of control
• variable on chromosome-like structure
• search through a population of points, not
  single point
• a fitness value is assigned
• to each individual, expressing
• its quality measure
• genetic operators are applied in order to create
  new offspring from best fitted individuals

• Two-level approach
• Outer level search for the optimum process
  time, with a suitable algorithm
• Genetic algorithms
• Luus Jaakola
• Hooke Jeeves
• Inner level search the optimum command variable
  profile such as for the fixed end-time case

Solution strategy
Modified Denbigh problem free end-time
10 Time-slices Performance index 0.543 Process
time tf 467.9 s
30 Time-slices Performance index 0.547 Process
time tf 526.9 s
45 Time-slices Performance index 0.5474 Process
time tf 800 s
60 Time-slices Performance index 0.5458 Process
time tf 800 s

• Conclusions
• GA technique was implemented for computation of
  piece-wise optimal control variable profiles for
  free end-time cases (two-level approach)
• GA performs slightly better than other current
  techniques (dynamic programming, iterative
  dynamic programming)
• Computation time significantly larger
• Optimal GA parameters should be find

90 Time-slices Performance index 0.5435 Process
time tf 800 s
15 Time-slices Performance index 0.5423 Process
time tf 640.3 s