基于多机器学习算法耦合的空气质量数值预报订正方法研究及应用

Research and Application of an Ensemble Forecasting Method Based on Coupled Multi-Machine Learning Algorithms

应用多种机器学习算法进行时空耦合从而建立一种新的多模式集合预报订正算法(简称“ET-BPNN算法”),对4种常规污染物(NO_(2)、O_(3)、PM_(2.5)和PM_(10))的空气质量模型预报结果进行订正.订正方法分为两步,第一步中利用随机森林、极端随机树和梯度提升回归树3种机器学习算法,采用4个空气质量数值预报模式(CMAQ、CAMx、NAQPMS和WRFChem)的多尺度污染物浓度预报数据、中尺度天气模式(WRF)的气象因子预报数据(包括2 m温度、2 m相对湿度、10 m风速、10 m风向、气压和小时累计降水量)以及污染物浓度观测数据作为训练集,训练结果进入基于均方根误差的择优选择器,选取3种机器学习算法中优化效果最好的算法;在第二步中利用了BP神经网络算法,通过加权平均获得集合模式订正预报结果.结果表明:①与模式集合平均算法相比,ET-BPNN算法使NO_(2)、O_(3)、PM_(2.5)和PM_(10)浓度预报值与观测值之间的均方根误差分别减小了30.4%、18.9%、43.3%和38.1%.②ET-BPNN算法的优化效果较随机森林、极端随机树和梯度提升回归树3个机器学习算法有明显提升,与极端随机树算法相比,ET-BPNN算法使NO_(2)、O_(3)、PM_(2.5)和PM_(10)浓度预报值与观测值之间的均方根误差分别降低了42.7%、20.1%、19.7%和9.7%.③在易发生污染的秋冬季,ET-BPNN算法对PM_(2.5)浓度的预报具有明显的优化效果,此外该算法明显缩小了不同站点预报和不同预报时效之间的偏差,具有较好的鲁棒性.④对O_(3)和PM_(2.5)浓度预报而言,经ET-BPNN算法优化后的预报结果能够更好地把握污染过程,对污染物峰值浓度的预报也较模式集合平均算法更准确.研究显示,ET-BPNN算法提高了空气质量模式对污染物浓度的预报效果.

In order to optimize and evaluate the air quality forecast accuracy of four pollutants(NO_(2),O_(3),PM_(2.5) and PM_(10))in Shanghai,a new spatial-temporal coupling ensemble forecasting method(ET-BPNN,short for ensemble tree-back propagation neural network)was established based on four machine learning algorithms.Firstly,the forecasting results were optimized using random forest,extreme random tree and gradient boosting decision tree,with input features selected from multi-scale forecasting data based on four air quality numerical forecasting models(CMAQ,CAMx,NAQPMS and WRFChem),the meteorological data of mesoscale weather model(WRF,including 2 m temperature,2 m humidity,10 m wind speed,10 m wind direction,atmospheric pressure and hourly accumulated precipitation)and observations.The best machine learning algorithm was chosen by comparing the root mean square errors.Secondly,further optimization was proceeded with BP neural network.The results show that:(1)Compared with the traditional ensemble mean algorithm of four air quality numerical forecasting models,the root mean square error(RMSE)between the ET-BPNN simulation and observed hourly concentration of NO_(2),O_(3),PM_(2.5) and PM_(10) was reduced by 30.4%,18.9%,43.3%and 38.1%,respectively.(2)The optimization effect of the ET-BPNN algorithm was significantly improved compared with the three machine learning algorithms,and the RMSE of NO_(2),O_(3),PM_(2.5) and PM_(10) were reduced by 42.7%,20.1%,19.7%and 9.7%,respectively.(3)ET-BPNN had an good optimization effect on PM_(2.5) forecasting in autumn and winter when its concentration was high,and the forecast bias of different stations was reduced.(4)The ET-BPNN also improved the forecast performance for pollution process of O_(3)-8 h and PM_(2.5),with peak value simulated more accurate than the traditional ensemble average algorithm.The study shows that the ET-BPNN algorithm improves the prediction effect of air quality models.