计算机工程与应用 ›› 2018, Vol. 54 ›› Issue (19): 107-113.DOI: 10.3778/j.issn.1002-8331.1706-0206

• 模式识别与人工智能 • 上一篇    下一篇

改进的粒子群算法优化神经网络及应用

何明慧1,徐  怡1,2,王  冉1,胡善忠1   

  1. 1.安徽大学 计算机科学与技术学院,合肥 230601
    2.安徽大学教育部智能计算与信号处理实验室,合肥 230039
  • 出版日期:2018-10-01 发布日期:2018-10-19

Combination dynamic inertia weight particle swarm optimization algorithm to optimize neural network and application

HE Minghui1, XU Yi1,2, WANG Ran1, HU Shanzhong1   

  1. 1.College of Computer Science and Technology, Anhui University, Hefei 230601, China
    2.Key Laboratory of Intelligent Computing and Signal Processing, Ministry of Education, Anhui University, Hefei 230039, China
  • Online:2018-10-01 Published:2018-10-19

摘要: 针对神经网络权值选取不精确的问题,提出改进的粒子群优化算法结合BP神经网络动态选取权值的方法。在改进的粒子群优化算法中,采用动态惯性权重,并且认知参数与社会参数相互制约。同时,改进的粒子群优化算法结合差分进化算法使粒子拥有变异与交叉操作,保持粒子的多样性。基于改进的粒子群优化算法与BP神经网络,构建IPSONN神经网络模型并运用于酒类品质的预测。实验分别从训练精度、正确率及粒子多样性三方面验证了IPSONN模型的有效性。

关键词: 神经网络权值, 粒子群优化算法, 动态惯性权重, 变异与交叉, 有效性

Abstract: In view of inaccurately selecting problem of neural networks weights, the method of improved particle swarm optimization algorithm combined with BP neural network to select weights dynamically is proposed. The improved particle swarm optimization algorithm applied dynamic inertia weight, cognitive parameter and social parameter is restrained each other. Simultaneously, the improved particle swarm optimization algorithm combined with differential evolution algorithm to make particles have mutation and crossover operation, maintain the particle diversity. The IPSONN neural network model based on the improved particle swarm optimization algorithm and BP neural network is constructed. The model is applied in forecasting the wine quality. Experiments prove the effectiveness of the IPSONN model from three aspects:training accuracy, correct rate and particle diversity.

Key words: neural network weights, particle swarm optimization algorithm, dynamic inertia weight, mutation and crossover, effectiveness