|
| 基于RBF神经网络的PM2.5浓度预测 |
| Prediction of PM2.5 Concentration Based on RBF Neural Network |
| |
| DOI: |
| 中文关键词: PM2.5 RBF神经网络 粒子群算法 大气污染物 气象因素 回归预测 |
| 英文关键词: PM2.5 RBF neural network Particle swarm optimization Air pollutants Meteorological factors Regression prediction |
| 基金项目:国家自然科学基金资助项目(41864002) |
|
| 摘要点击次数: 306 |
| 全文下载次数: 241 |
| 中文摘要: |
| 针对传统RBF神经网络在PM25回归预测中参数优化的问题,提出了粒子群算法优化的径向基神经网络(PSO-RBF)、鲸鱼算法优化的径向基神经网络(WOA-RBF)、北方苍鹰算法优化的径向基神经网络(NGO-RBF)和灰狼算法优化的径向基神经网络(GWO-RBF)4种模型,以2021年12月1日—2022年8月31日拉萨、成都、北京和上海的大气污染物、气象因素、大气可降水量(PWV)及叶面积指数(LAI)的小时数据作为训练集,分别预测了4个城市在2022年9月、10月、11月共计91 d的PM25质量浓度变化。结果表明:PSO-RBF模型的优化性能最为显著,相对于RBF模型,PSO-RBF模型的MAE、MAPE、RMSE、R2均得到显著提升。 |
| 英文摘要: |
| Regarding the problem of parameter optimization of traditional RBF neural network in PM25 regression prediction, this paper proposed four radial basis function network models optimized by different algorithms, namely particle swarm optimization(PSO RBF),whale optimization algorithm(WOA RBF), northern goshawk optimization(NGO RBF) and grey wolf optimizer(GWO RBF). Taking the hourly data of air pollutants, meteorological factors, precipitable water vapor(PWV) and leaf area index(LAI)in Lhasa, Chengdu, Beijing and Shanghai from December 1st, 2021 to August 31st, 2022 as the training set, the changes of PM25 mass concentration for a total of 91 days in September, October and November 2022 in the four cities were predicted respectively. Results showed that the optimization performance of the PSO RBF model was the most significant. Compared with the RBF model, the MAE, MAPE, RMSE and R2 of the PSO RBF model have been significantly improved. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |