基于遗传算法和神经网络的复杂对象的建模与优化

针对复杂对象全局优化结果较多,难以选取最佳参数组合的问题,提出了一种新适应度函数,其特点是使用了目标函数关于其影响参数的梯度之和。结合提出的新适应度函数,首先进行已知模型的遗传算法寻优,然后对神经网络构建的模型进行遗传算法寻优。两次试验结果都表明,搜索到的最佳性能目标对应的参数组合具有较强的鲁棒性。

Modeling and Optimization of Complex Objects Based on Genetic Algorithm and Neural Network

There are many global optimization results for complex objects,so it is hard to select the best parameters,this paper proposes a new fitness function with the sum of gradients of the parameters on the original objective function.At first optimized known model with genetic algorithm and the new fitness function,and then built the model with neural network and did the optimization.The results show that the best parameters have strong robustness.