A mixed-coding adaptive differential evolution for optimising the architecture and parameters of feedforward neural networks

Abstract

This paper presents an adaptive differential evolution with mixed-coding strategy to evolve feedforward neural networks (FNNs). This algorithm with adaptive control parameters which can handle effectively binary variables and real variables, is used to optimise simultaneously FNN architecture and connection parameters (weights and biases) by a specific individual representation and evolutionary scheme. The performance of the algorithm has been evaluated on several benchmarks. The results demonstrate that the proposed algorithm can produce compact FNNs with good generalisation ability.