Real or Binary Coding

In any genetic algorithm application, the physical problem must be translated into a form suitable for manipulation by the evolutionary operators. Choice of coding is an important part of this process. 

In early work, GA strings were binary coded. Computer scientists are comfortable with binary representations and the problems tackled at that time could be easily expressed using this type of coding. Binary coding is sometimes appropriate in scientific applications, but it is less easy to interpret than alternative forms, as most scientific problems are naturally expressed using real numbers. 

Applying one-point crossover at a randomly chosen position, the thirty-second binary digit in each string, we get the two new strings  

When these new binary strings are translated back into real values so that we can check what has happened, the result is 

We could apparently get around this difficulty by allowing changes to be made to not just one but every bit in the mutation of the 9-bit binary string, but this then again returns us to a situation in which we are, in essence, manipulating real numbers disguised in a binary cloak. 

---