针对模块化神经网络的重要命题——子网动态集成问题,提出一种基于改进的Bayes学习的子网集结新方法.首先从处理复杂问题能力、计算开销、训练误差限等级的合理性、逼近正确率的构造等方面分析了已有方法的不足.既而提出相应策略,其核心在于采用了简洁、相关性小的子网生成方法;同时以误差作为依据提出新的逼近正确率指标以确定子网的动态集结权值.仿真实验对两种改进方法的测试误差进行了比较研究,结果表明了改进方法的有效性. Aiming at the problem of subnet dynamic integration, an important proposition of modular neural network, this paper proposes a new subnet aggregation method based on improved Bayesian learning.Firstly, from the aspects of the ability to deal with complex problems, the computational cost, the reasonableness of training error limits, The correctness of the structure and other aspects of the analysis of the shortcomings of the existing methods.Then put forward the corresponding strategy, the core lies in the use of a simple, low correlation subnet generation method; the same time error as a basis for the new approximation accuracy index to determine the child The dynamic aggregation weights of the network are compared.The simulation experiment is used to compare the test error of the two improved methods and the results show the effectiveness of the improved method.