论文部分内容阅读
在人工神经网络应用中,由于存在网络规模和拓扑结构难以预先确定,网络学习速度慢,且易于收敛到局部最优点等问题,有关文献提出了采用基于遗传算法(GAs) 思想进行设计和学习的方法,该方法能够同时确定网络的结构及有关参数。该文在此基础上,对此方法进行了改进,改进之处在于,采用浮点数矩阵来表示编码,同时对于遗传算法的进化过程也进行了一定的改进,使该方法能够接受一定的约束条件。针对前馈型神经网络,该方法在满足一定约束条件的情况下,能同时有效地寻找到合适的网络结构和相应的参数( 神经网络的权值和阈值) , 新方法较原方法在精度和速度上都有较大的提高。
In the application of artificial neural network, due to the existence of network size and topology is difficult to be pre-determined, the network learning speed is slow, and easy to converge to the local optimal point and other issues, the literature proposed the use of genetic algorithms (GAs) to design and learn Method, which can simultaneously determine the network structure and related parameters. On the basis of this, this method is improved. The improvement is that using floating-point matrix to represent coding, and the evolutionary process of genetic algorithm is also improved to some extent, so that the method can accept certain constraints . For the feedforward neural network, the proposed method can find the appropriate network structure and the corresponding parameters (the weights and thresholds of the neural network) at the same time under certain constraint conditions. Compared with the original method, Speed has greatly improved.