摘要
针对粒子群算法易陷入局部最优和寻优精度比较低等缺点,提出一种基于随机惯性权重和异步变化策略的学习因子的粒子群算法优化神经网络连接权重和阈值,并以此建立月降水预报建模研究.以广西桂北地区的月降水量实例分析,并与标准粒子群优化神经网络模型、随机权重的粒子群神经网络模型和神经网络模型对比,结果表明,该方法学习能力强和预测精度高,是一种有效的建模预报方法.
As the particle swarm algorithm is easy to fall into local optimum and the disadvantages of optimization low accuracy, this paper puts forward the evolving neural network connection weights and thresholds（or bias） based on random inertia weight and asynchro- nous changes＇ leaning factors of particle swarm algorithm （PSO）, which establish monthly rainfall forecasting mode. The applied exanlple analysis is built with the monthly rainfall in the area of north of Guilin in Guangxi, the new method has been compared with others forecast- ing models, such as SPSO-TFNN and TFNN etc. The experimental results show the presented approach has strong learning ability and high generalization performance in rainfall forecasting, and is an effective tool for runoff forecasting.
出处
《柳州师专学报》
2014年第1期126-130,共5页
Journal of Liuzhou Teachers College
基金
广西高校科学技术研究项目(2013YB281)
柳州师范高等专科学校基金资助项目(LSZ2012B005)
关键词
粒子群优化算法
人工神经网络
月降水预报
Particle Swarm Optimization
Artificial Neural Network
Monthly Runoff Forecasting
