基于粒子群BP神经网络的结冰大气参数预测  

Atmosphere Icing Parameters Prediction Based on PSO-BPNN

作　　者：郭琪磊 桑为民[1,2] 牛俊杰 Guo Qilei;Sang Weimin;Niu Junjie(School of Aeronautics,Northwestern Polytechnical University,Xi’an 710072,China;Key Laboratory of Icing and Anti/De-icing,China Aerodynamics Research and Development Center,Mianyang 621000,China)

机构地区：[1]西北工业大学航空学院,陕西西安710072 [2]中国空气动力研究与发展中心结冰与防除冰重点实验室,四川绵阳621000

出　　处：《气动研究与试验》2025年第2期96-102,共7页Aerodynamic Research & Experiment

基　　金：结冰与防除冰重点实验室开放课题(2001IADL20200101)。

摘　　要：准确预测结冰大气参数可为飞机防除冰策略制定与优化提供依据。以数值天气预报(WRF)模式为基础进行结冰大气参数预测,为提高WRF模式预测精度,采用粒子群优化(PSO)算法改进BP神经网络(BPNN),进而构建基于粒子群BP神经网络的结冰大气参数预测模型。以海南某地气象环境为预测对象,考察PSO-BPNN算法预测结冰大气参数的能力。结果表明,WRF模式预测的压力与温度与实际观测值较为一致,但仍存在偏差,需进一步修正方可满足实际需求;PSO-BPNN算法可显著提高WRF模式预测结冰大气参数的精度,绝对平均误差、均方根误差、绝对平均百分比误差均有较为明显降低,无论在时序变化还是数值精度上均与实测值更为吻合。此外,后续研究中将研究不同结冰大气参数间相关性,以期提高该算法修正的准确度。Accurate prediction of icing atmospheric parameters can pave the foundation for aircraft de-/anti-icing strategies.Based on numerical weather prediction(WRF)pattern,icing atmospheric parameters were predicted,and in order to improve the prediction accuracy,the particle swarm optimization algorithm was employed to optimize the BP neural network.Finally,a prediction model of icing atmospheric parameters based on PSO-BPNN was constructed.Considering the meteorological environment of Hainan province,the ability of PSO-BPNN to predict icing atmospheric parameters was investigated.The results showed that although the pressure and temperature predicted by the WRF pattern were relatively consistent with the actual observed values,there were still some deviations,which need to be further corrected.The PSOBPNN algorithm could significantly improve the accuracy of the predicted icing atmospheric parameters by the WRF pattern.MAE,RMSE and MAPE had been evidently reduced,and corrections were closer to the observations in terms of time series changes and numerical accuracy.In addition,in the follow-up research,the correlation among various icing atmospheric parameters will be investigated in order to improve the accuracy of algorithm correction.

关 键 词：大气参数预测 粒子群算法 BP神经网络 WRF 优化 

分 类 号：V211.3[航空宇航科学与技术—航空宇航推进理论与工程] P435.1[天文地球—大气科学及气象学]