| 多隐层BP神经网络在模式预报中的简化应用 |
| 点此下载全文 |
| 引用本文：沈沉，赵文灿，施金海，等..多隐层BP神经网络在模式预报中的简化应用[J].气象与环境科学,2019,42(4):127-132.Shen Chen,Zhao Wencan,Shi Jinhai,et al..Simplified Application of Multi Layer BP Neural Network in Model Prediction[J].Meteorological and Environmental Sciences,2019,42(4):127-132. |
| 摘要点击次数: 426 |
| 全文下载次数: 285 |
|
| DOI：doi:10.16765/j.cnki.1673-7148.2019.04.018 |
| 基金项目: |
|
| 中文摘要:EC细网格预报效果好，基本满足业务需要，在工作中被广泛应用。为进一步提升预报准确性，做好迪士尼园区的气象服务保障，选取2016年7月至2017年6月1年的2 m温度预报场，24 h预报时效的时间分辨率为3 h，72 h预报时效的时间分辨率为24 h，分别用回归分析法、S型和简化Line型BP神经网络法进行模式释用，与迪士尼气象站观测数据对比。结果表明：阈值为1 ℃时，对模式结果释用后，均方根误差减少了0.5 ℃到1.0 ℃，39 h和2172 h预报时效的准确率由原来的50%和30%分别上升到70%和50%。采用S型多隐层BP神经网络误差最小，不同预报时效释用稳定性最高，同时该释用方法对tmin的预报特征把握更精准，释用效果明显优于对tmax的预报释用，但迭代计算耗费时间大幅增多，与预报效果的提升不成正比。简化Line型的BP神经网络通过8个半月的数据量和简单的网络模式，捕获了EC预报的特征，不但减小了计算量，大幅缩短了计算时间，而且预报结果也有显著提升，预报稳定性较好，具有广泛的业务应用空间。 |
| 中文关键词:上海迪士尼度假区 地面气温 BP神经网络 模式释用 |
| |
| Simplified Application of Multi Layer BP Neural Network in Model Prediction |
|
|
| Abstract:Fine grid EC forecast has good effect,basically meets business requirement and is widely used in work.In order to improve prediction accuracy and ensure the meteorological services in Disney resort,the forecast data of 2 m temperature with resolution of 3 h in 24 h and 24 h in 72 h forecast period from July 2016 to June 2017 are chose,at the same time,the regression analysis,S type,simplified Line type and BP neural network method are used to interpret the model,and compared with surface data from Disney meteorological station.The results are as follows.When threshold is 1 ℃,after model interpretation,root mean square has been reduced by 0.5 ℃ to 1.0 ℃,forecast accuracies in 39 h and 2172 h are greatly improved from 50% and 30% to 70% and 50% respectively.Using S type multi layer BP neural network the error is the smallest and interpretation is the most stable,what’s more,this method is more accurate for tmin prediction and its result is better than that of tmax prediction,however the time of iterative calculation is greatly increased,which is not directly proportional to the improvement of prediction effect.Simplified Line type BP neural network captures the characteristics of EC forecast through over eight months data and simple network mode,not only reduces computational quantity and greatly shortens the time,but also improves forecast results significantly and keeps stable forecast,which has extensive business application space. |
| Keywords: |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|