Real-value encoding represents each solution parameter directly as a real-valued gene rather than as a binary or Gray-coded string. It fits genetic algorithms whose decision variables are continuous or naturally expressed as real-valued parameters. section 8.2 section 7.3.2
Choose real-value encoding when the optimization problem benefits from:
- Continuous precision without discretizing the search space into a finite set of binary values. section 8.2
- More compact representation and potentially lower memory and computation requirements than high-precision binary encoding. section 8.2
- Smoother locality, reducing the disruptive jumps associated with binary Hamming cliffs. section 8.1 section 8.9
- Direct treatment of curve-fitting variables, real-valued function inputs, PID controller parameters, or neural-network weights. section 8.2
Precision without a long bit string
Suppose a variable ranges from 0 to 10 and requires precision of 0.0001. Binary encoding needs 17 bits for the required range and precision; real-value encoding can represent the parameter directly as a floating-point value, preserving continuous search while avoiding that discretization. section 8.2
