【作者】 王睿；

【导师】 李银林；

【作者基本信息】 北京理工大学 ， 兵器科学与技术， 2018， 硕士

【摘要】 本文在简要介绍国内外精细化天气预报水平和数据挖掘在气象领域的应用现状基础上,展开了基于机器学习和数据挖掘的精细化温度预报方法研究。收集整理了我国北方地区三个站点的温度历史实况资料,针对时间序列数据可以分解为趋势分量和波动分量的性质,设计实现了利用深度递归神经网络算法和启发式免疫算法分别对温度的趋势项和季节性波动项进行拟合后再叠加的预测模型,并在MATLAB平台上进行仿真,利用数据集中前29年的数据进行网络训练,以最后一年的数据进行检验验证。通过与历史实况的对比,分析了模型预报的准确率和稳定性,证明本文提出的精细化温度预报算法合理可行,可以提高现有预报水平,对温度预报的辅助决策具有重要参考意义。更多还原

【Abstract】 In this paper,a refined temperature forecasting method based on machine learning and data mining is developed based on a brief introduction of refined weather forecasting and the application status of data mining in meteorological field.The temperature historical observation data of three stations in northern China were collected and arranged.The time series data can be decomposed into the nature of the trend component and the fluctuation component.The trend of the temperature of the recursive neural network algorithm and the heuristic immune algorithm are respectively designed and implemented.The prediction models that were fitted after seasonal fluctuations were fitted and simulated on the MATLAB platform.The data of the previous 29 years of data set were trained on the network,and the final year’s data was used for verification.By comparing with the historical facts,the accuracy and stability of the model forecasting are analyzed.It is proved that the refined temperature forecasting algorithm proposed in this paper is reasonable and feasible.It can improve the existing forecasting level and has important reference significance for the auxiliary decision-making of temperature forecasting.更多还原