A study on scalable representations for evolutionary optimization of ground structures.

Area of Science:

• Computational Engineering
• Artificial Intelligence
• Evolutionary Computation

Background:

• Indirect solution representations are crucial for complex design optimization problems.
• Generative and ontogenic representations offer distinct approaches to encoding design parameters.
• Evaluating these representations requires assessing both solution quality and computational efficiency.

Purpose of the Study:

• To comparatively analyze generative and ontogenic indirect solution representations for 2D truss design.
• To evaluate the performance of these representations using evolutionary strategies and HyperNEAT.
• To assess the impact of representation choice on solution optimality and computational cost.

Main Methods:

• Implementation of generative and ontogenic representations for 2D truss design.
• Application of a naive evolution strategy and the HyperNEAT algorithm for optimization.
• Comparative analysis based on objective values and computational expense.

Main Results:

• The ontogenic representation consistently yielded near-optimal solutions for 2D truss design.
• The generative representation produced results far from optimal under similar conditions.
• The ontogenic representation demonstrated significantly lower computational cost until near-optimal levels were achieved.

Conclusions:

• The ontogenic representation is highly effective and computationally efficient for 2D truss design optimization.
• Generative representations raise concerns regarding scalability in complex design tasks.
• Ontogenic representations exhibit superior scalability and performance in evolutionary design.