摘要
针对传统的BP神经网络收敛速度较慢且易陷入局部最优解的问题,文章提出了一种基于粒子群优化(PSO)算法优化BP神经网络的PM_(2.5)浓度预测模型,从而能够快速收敛并得到全局最优解。首先,通过皮尔逊相关性分析筛选出与PM_(2.5)浓度相关性较高的污染物指标作为输入变量。其次,利用PSO算法优化BP神经网络的初始权重和阈值,克服了BP神经网络易陷入局部最优、收敛速度慢的缺点。最后,利用成都市2021年7月至2024年6月的大气污染物数据对模型进行训练和测试。结果表明,测试集的R^(2)达到0.944,测试集的MAE为4.231,测试集的RMSE为6.364。与未优化的BP神经网络模型相比,PSO-BP模型具有更高的预测精度和更快的收敛速度,能够有效地预测成都市次日的PM_(2.5)浓度。
Aiming at the problem that the traditional BP Neural Network has slow convergence speed and is easy to fall into local optimal solution,this paper proposes a PM_(2.5)concentration prediction model based on Particle Swarm Optimization(PSO)algorithm optimized BP Neural Network,which can quickly converge and get the global optimal solution.Firstly,the pollutant indexes with high correlation with PM_(2.5)concentration are selected as input variables by Pearson correlation analysis.Secondly,the PSO algorithm is used to optimize the initial weights and thresholds of BP Neural Network,which overcomes the shortcomings of BP Neural Network,such as easy to fall into local optimum and slow convergence speed.Finally,the model is trained and tested using air pollutant data from July 2021 to June 2024 in Chengdu.The results show that the R^(2)of the test set is 0.944,the MAE of the test set is 4.231,and the RMSE of the test set is 6.364.Compared with the unoptimized BP Neural Network model,the PSO-BP model has higher prediction accuracy and faster convergence speed,and can effectively predict the PM_(2.5)concentration of the next day in Chengdu.
作者
李佳林
侯利明
张聪
LI Jialin;HOU Liming;ZHANG Cong(Sichuan Vocational College of Health and Rehabilitation,Zigong 643000,China;Xinxiang Medical University,Xinxiang 453003,China)
出处
《现代信息科技》
2025年第7期47-51,57,共6页
Modern Information Technology
基金
四川卫生康复职业学院重点课题(CWKY-2019Z-02)
四川卫生康复职业学院校级科研团队(CWKY-TD24-10)。
作者简介
李佳林(1987-),男,汉族,四川内江人,讲师,博士在读,研究方向:智能计算和深度学习;侯利明(1987-),男,汉族,河南新乡人,副教授,博士,研究方向:深度学习;张聪(1992-),男,汉族,四川自贡人,工程师,硕士在读,研究方向:机器学习。
