摘要
为能更有针对性的控制PM2.5浓度,对2000-2017年间全国31个省市PM2.5浓度数值和由专家先验得出的影响PM2.5的六种人为因素分别建立了基于BIC评分函数、K2评分函数进行结构学习的两种贝叶斯网络模型、支持向量机模型、K-近邻模型进行分析。利用5-折交叉验证对四种模型进行评估。发现在样本量不太大的情况下,贝叶斯网络表现出更好地稳健性与优越性,而基于K2评分函数进行结构学习的贝叶斯网络模型具有更好地分类性能。为政府相关部门对我国PM2.5浓度更加有效的控制,以及采取更加具有针对性的治理方案提供了思路与方案。
For controlling the concentration of the PM2.5,with the data of PM2.5concentration values of 31 provinces in China during theyears 2000 to 2017,and six influencing factors of PM2.5concentration obtained by experts’prior experience.Two Bayesian Network Models based on BIC score function and K2 score function respectively,Support Vector Machine,K-Nearest Neighbor Model are analyzed.Using the method of 5-fold cross validation,the four models are trained and tested.In conclusion,in the case of small sample size,the Bayesian Network shows better robustness and superiority,while the Bayesian Network Model based on K2 score function for structural learning has better performance.It is meaningful to provide a new idea and scheme for the control of PM2.5concentration in China.
出处
《科技创新与应用》
2020年第22期1-5,9,共6页
Technology Innovation and Application
基金
国家社会科学基金重大招标项目(编号:18ZDA052)
上海市科委科研计划资助项目(编号:14DZ2280200)。
关键词
PM2.5
贝叶斯网络
结构学习
参数学习
模型评估
PM2.5
bayesian network
structure learning
parameter learning
assessment of models