摘要
针对水质预测中传统BP神经网络模型收敛速度慢,对隐层结点选取缺乏有效的手段等问题,引入了遗传算法优化BP网络的结构和隐层神经元阈值和连接权值,通过设计灵活的实数编码方案和新型交叉算子等,对实数编码遗传算法进行改进,在此基础上,提出了一种基于改进的实数编码遗传算法优化BP神经网络(IGA-BP)的水质预测新模型,并以安徽蚌埠蚌埠闸逐周水质监测的PH值数据为例,进行水质预测,通过与传统的GA-BP神经网络以及BP神经网络的水质预测模型对比,结果表明,这种预测方法训练的BP神经网络收敛速度快,样本逼近精度高且泛化能力强。
There are a number of problems associated with water quality prediction when using the traditional BP neural network model,such as slow convergence and lack of efficient methods for selecting hidden layer neural nodes. This study introduces a genetic algorithm to optimize the structure of the BP network,the thresholds,and the connection weights of hidden layer neural nodes. The real-coded genetic algorithm was optimized by redesigning the flexible real-coded scheme and introducing a new crossover operator. A new water quality prediction model based on the improved genetic algorithm（ IGA-BP） was proposed to optimize the BP neural network,and was tested using the weekly monitored water quality data at Bengbu water gate in Anhui Bengbu. The output was compared to the water quality prediction model based on the traditional GA-BP and BP neural networks. The results showed that this newly proposed water quality prediction method using IGA-BpHad a faster convergence speed,higher sample approximation accuracy,and stronger generalization abilities.
出处
《环境工程学报》
CAS
CSCD
北大核心
2016年第3期1566-1571,共6页
Chinese Journal of Environmental Engineering
基金
上海市科学技术委员会科技创新项目(12595810200)
上海海事大学科研项目
关键词
水质预测
BP神经网络
实数编码遗传算法
优化
交叉操作
water quality prediction
BP neural network
real-coded genetic algorithm
optimization
crossover operator
