基于面向对象自适应粒子群算法的神经网络训练*

针对传统的神经网络训练算法收敛速度慢和泛化性能低的缺陷，提出一种新的基于面向对象的自适应粒子群优化算法(OAPSO)用于神经网络的训练。该算法通过改进PSO的编码方式和自适应搜索策略以提高网络的训练速度与泛化性能，并结合Iris和Ionosphere分类数据集进行测试。实验结果表明：基于OAPSO算法训练的神经网络在分类准确率上明显优于BP算法及标准PSO算法，极大地提高了网络泛化能力和优化效果，具有快速全局收敛的性能。

Neural network training based on object-oriented adaptive particle swarm optimization

In view of the traditional neural network training algorithm defects of slow convergence speed and the low generalization, this paper proposed a novel object-oriented adaptive particle swarm optimization(OAPSO) algorithm in the neural network training. This algorithm enhanced the training speed and the generalization of network through improving the encoding method and the self-adapted search strategy of PSO. Then, used two standard data sets, Iris and Ionosphere, in the test. The experiments show that the neural network based on OAPSO algorithm is obviously superior to BP algorithm and standard PSO algorithm in the classification accuracy rate, and enhances the generalization and the optimized effect of the network. This algorithm has the performance of rapid global convergence.