改进粒子群优化BP神经网络的PM2.5预测  

PM2.5 forecasting based on improved particle swarm optimizing BP neural network

作　　者：贾佳美 池凯凯[1] 吴哲翔 JIA Jia-mei;CHI Kai-kai;WU Zhe-xiang(School of Computer Science and Technology,Zhejiang University of Technology,Hangzhou 310000,China;Jinhua State Grid,Power Supply Company of Zhejiang,Jinhua 321000,China)

机构地区：[1]浙江工业大学计算机科学与技术学院,浙江杭州310000 [2]国网浙江省电力公司金华供电公司,浙江金华321000

出　　处：《计算机工程与设计》2021年第12期3495-3501,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(61872322)。

摘　　要：针对PM2.5预测的非线性不确定特点,提出基于改进粒子群优化BP神经网络的空气PM2.5浓度预测模型。引入混沌映射和对立学习改进粒子群算法;引入对立学习提高初始解的质量;引入混沌Tent映射改进粒子随机搜索,避免局部最优;引入自适应惯性权重均衡局部开发和全局勘探能力。利用改进粒子群对BP神经网络权值和阈值进行迭代寻优,基于最优参数BP神经网络做PM2.5预测,有效避免神经网络训练时陷入局部最优,提升收敛速度。选取某市某时段的PM2.5日均浓度数据进行实验分析,结果表明IPSO-BP预测准确度更高,收敛速度更快。Aiming at nonlinear and uncertain characteristics in PM2.5 forecasting,a PM2.5 prediction model combined with improved particle swarm optimization for BP neural network was designed.Particle swarm optimization was improved by introducing chaotic mapping and opposition-learning.The opposite-learning was introduced for improving the quality of initial solution.A chaos Tent mapping was introduced to improve the random searching of particles,which avoided falling into local optimal.A self-adaptive inertia weight was designed to trade off the ability of local exploration and global exploration.The proposed improved PSO was applied to search the optimal parameters of BP neural network in the iterative way,such as weights and bias.BP neural network with optimal parameters was applied to forecast PM2.5.The designed method could refrain from a local optimum and enhance the searching convergence rate when the neural network was trained.A daily average PM2.5 concentration in a certain period in certain city was selected as experimental analysis data.Results indicate that the proposed algorithm IPSO-BP has higher prediction precision and convergence rate.

关 键 词：PM2.5预测 BP神经网络 粒子群优化算法 混沌Tent映射 对立学习 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]