论文部分内容阅读
利用智能体的思想,设计多个agent进行了多层前馈神经网络的协作训练,它可以克服只用遗传算法训练神经网络时的低效率或只用BP算法进行神经网络训练时容易陷入局部极小点等的缺限.对比试验结果表明,这种新的训练形式不仅能够提高收敛速度,避免局部最小点,而且可以利用专家知识进行训练优化,提高系统的通用性
Using the idea of agent, this paper designs multiple agents and conducts cooperative training of multi-layer feedforward neural network. It can overcome the inefficiency of training neural network using only genetic algorithm or neural network training with BP algorithm. Lack of dots and so on. The results of comparative experiments show that this new training form can not only improve the convergence speed and avoid the local minimum, but also make use of expert knowledge to optimize training and improve the universality of the system