Combining with genetic algorithm, the improved Estimation of Distribution Algorithm (EDA) is provided. The crossover and mutation operations are added and the "elite" individuals are retained, which can keep the excellent evolution mode. The selection based on energy entropy is added, which can explore the solution space sufficiently and keep the population diversity. A neural network with switches introduced to its links is proposed. The method of tuning the structure and parameters of the neural network using the improved EDA is provided. The carrying robot inverse dynamics model approximation example show the validity of this algorithm. Traditional evolutionary algorithm achieves the population evolution based on the genetic manipulation for the every individual in the population (crossover and mutation, etc.) and establishes the mathematical model from the "micro" level. But EDAs directly describes the evolution trend of the population through the establishment of a mathematical model based on the whole population from the "macro" level in biological evolution. This paper combines the probabilistic manipulation for EDA in the "macro" level with the genetic manipulation in the "micro" level, and calculates the individual's information entropy by adding individual choice mechanism based on information entropy, thus increases the diversity of the population diversity. Combining the "elite" individual reservation mechanism, a fast parallel EDA with the global search capabilities is provided.
[1]
J. A. Lozano,et al.
Estimation of Distribution Algorithms: A New Tool for Evolutionary Computation
,
2001
.
[2]
David E. Goldberg,et al.
A Survey of Optimization by Building and Using Probabilistic Models
,
2002,
Comput. Optim. Appl..
[3]
David E. Goldberg,et al.
Linkage Problem, Distribution Estimation, and Bayesian Networks
,
2000,
Evolutionary Computation.
[4]
Pedro Larrañaga,et al.
Estimation of Distribution Algorithms
,
2002,
Genetic Algorithms and Evolutionary Computation.
[5]
Hak-Keung Lam,et al.
Tuning of the structure and parameters of a neural network using an improved genetic algorithm
,
2003,
IEEE Trans. Neural Networks.
[6]
Zhang Yi,et al.
Human-machine intelligent robot system control based on study algorithm
,
2008,
2008 IEEE International Conference on Industrial Technology.