摘要
月降水量变化呈现显著的非线性特征,对其进行精准预测难度很大。近年来,长短时记忆网络(LSTM)在降水量预测中优势明显。然而,LSTM的深层结构造成了其存在过拟合、时滞等缺点,从而影响预测精度。借助平行学习结构-宽度学习系统(BLS)直接计算权重的特点,提出改进的LSTM-BLS降水量预测模型。选取湖北省五个具有不同气候特征的代表性测站点进行实证研究。结果表明,与基线模型和已有的预测模型相比,现有模型在所有评价指标上均预测精度最高。特别地,由于BLS模块加入解决了LSTM的时滞性问题,新模型在强降水和干旱月份预测精度提升明显。不同时间步长下,新模型预测精度亦表现最佳,证明了其稳定性。在运算效率上,LSTM-BLS和LSTM相比,并未降低。
The inter-annual variability of monthly precipitation presents significant non-linear characteristics,making it very difficult to accurately predict.Recent years have witnessed the advantages of Long-Short Term Memory(LSTM)in precipitation prediction.However,deep structure of LSTM causes its shortcomings such as over-ftting and time lag,which affects the prediction accuracy.With the help of the parallel learning structure-Broad Learning System(BLS),which can directly calculate the weights,an improved LSTM-BLS precipitation prediction model is proposed.Five representative measurement sites with different climatic characteristics in Hubei Province were selected for empirical research.The results show that,compared with baseline and existed modes,all evaluation indicators of the new models have significantly improved.Especially,in the dry and heavy rain months,owing to the introduction of BLS solving the lag-time problem in LSTM,prediction accuracy has improved dramatically in the new model.Under different time steps,the prediction accuracy of the new model also performed best,proving its stability.And in terms of computational efficiency,LSTM-BLS has not decreased compared with LSTM.
作者
韩莹
谈昊然
王乐豪
罗嘉
HAN Ying;TAN Hao-ran;WANG Le-hao;LUO Jia(School of Automation,Nanjing University of Information Science and Technology,Nanjing Jiangsu 210044,China;Hubei Public Meteorological Service Center,Wuhan Hubei 430074,China)
出处
《计算机仿真》
北大核心
2023年第5期535-540,共6页
Computer Simulation
基金
南方海洋科学与工程广东省实验室(珠海)基金(SML2020SP007)
国家自然科学基金(62076136)。
关键词
降水量预测
长短时记忆网络
宽度学习系统
时滞
Precipitation prediction
Long and short-term memory network
Broad learning system
Time lag
