基于神经网络的混合优化算法研究

为了在资源有限的嵌入式系统中提高智能特性,需要研究开发具有较好逼近和泛化能力的、以及代码简单的神经网络.本文采用具有很强全局寻优能力的遗传算法以及局部寻优能力较强的梯度下降法,结合两者的优点形成了一种神经网络混合优化算法.对遗传算法操作算子做了改进,能同时优化神经网络的结构和权值及阈值.仿真实验结果表明,该混合优化算法能够有效地对神经网络的结构和权值及阈值优化,优化后的神经网络有较好的逼近能力和泛化能力.

Research of the Improved Hybrid Optimizational Algorithm on the Neural Network

In order to raise intelligent features in the embedded systems with limited resources,we need to research and develo Pwith a good approximation and generalization capabilities,as well as code for a simple neural network.In this article,we use a strong ability of overall optimization of genetic algorithm,local optimization ability and strong gradient descent method,combined with the advantages of both the formations of a hybrid optimization algorithm for neural networks.The genetic algorithm operators have m...