PSO-PSR-ELM集成学习算法在地面气温观测资料质量控制中的应用

Application of PSO-PSR-ELM-Based Ensemble Learning Algorithm in Quality Control of Surface Temperature Observations

针对台站密度低、新建台站以及特种单要素站等无法获得有效邻站或内部参考资料情况下的质量控制问题,从气温时间序列的混沌特性出发,考虑气温在短时间内的连续性和稳定性,提出一种基于粒子群(Particle Swarm Optimization,PSO)改进的相空间重构法(Phase Space Reconstruction,PSR)和极限学习机(Extreme Learning Machine,ELM)的集成学习算法的地面逐时气温观测资料的单站质量控制方法,实现气温资料的质量控制。为检验该方法的适用性,运用该方法对江苏省八市2007~2009年的地面气温观测资料进行质量控制,并与传统单站方法及切比雪夫多项式内插法(Tshebyshev Polynomial Interpolation,TPI)进行对比。实验结果表明,该方法相比较于TPI和传统方法可以更有效地标记出可疑数据,具有检错率高、地区和气候适应性、可控性强等优点。

In order to overcome the quality control problem over areas where the station density is low, or for some stations that have no adjacent stations and lack effective internal reference data, for example those newly deployed stations and special single factor stations, a new quality control method for surface temperature observations based on the ensemble learning algorithm of Phase Space Reconstruction（PSR） and Extreme Learning Machine（ELM） that was improved by Particle Swarm Optimization（PSO） was introduced in detail in this paper. This method considers the chaotic characteristics of the time series of temperature. In order to assess the feasibility and applicability of the proposed method, it was applied to hourly temperature observations from 2007 to 2009 in eight cities of Jiangsu Province. Results were examined and compared against that of conventional single-station quality control method and Tshebyshev Polynomial Interpolation（TPI） method. It was found that the method introduced in this study can flag suspicious data more effectively, and it also has the advantages of high identification accuracy and good adaptability and controllability over different regions with various climate backgrounds.