{"title":"基于 CEEMDAN-VMD-BiLSTM 网络的降雨预测模型","authors":"Sen Hou, Qikang Geng, Yaru Huang, Zhen Bian","doi":"10.1007/s11270-024-07299-8","DOIUrl":null,"url":null,"abstract":"<p>Rainfall prediction, based on meteorological data and models, forecasts the possible rainfall conditions for a period in the future. It is one of the important issues in meteorology and hydrology, and holds significant scientific and social value for enhancing human society's adaptive capacity, reducing the risk of natural disasters, promoting sustainable development, and protecting the environment. This study proposes a rainfall prediction model based on CEEMDAN-VMD-BiLSTM, which couples CEEMDAN (Complete Ensemble Empirical Mode Decomposition with Adaptive Noise), VMD (Variational Mode Decomposition), and BiLSTM (Bidirectional Long Short-Term Memory). The model first employs CEEMDAN and VMD, two decomposition algorithms, for a secondary decomposition of the original data, followed by prediction using the BiLSTM network. The study integrates the characteristics of CEEMDAN, which include adaptability, completeness, denoising capability, and high precision, the characteristic of VMD in extracting trend information, and the ability of the BiLSTM model to better capture contextual information in sequence data and solve long-term dependency issues, thereby increasing the accuracy of rainfall prediction. The research selected Zhongwei City in the Ningxia Hui Autonomous Region as the study object and used 20 years of monthly rainfall data from 2001 to 2020 as the research data. The model was compared with standalone BiLSTM models, CEEMDAN-BiLSTM coupled models, and VMD-BiLSTM coupled models. The model was validated using four indicators: RMSE, MARE, MAE, and NSE. The results showed that the maximum relative error of the CEEMDAN-VMD-BiLSTM neural network rainfall prediction coupled model was 7.22%, and the minimum relative error was -7.03%. The prediction qualification rate was 100%. The overall NSE value of the model ranged from 0.63 to 0.97, with most values between 0.86 and 0.97. The excellent rate was about 84.6%, and the good and above rate was 92.3%. In the rainfall prediction for Zhongwei City, the prediction accuracy of this coupled model was better than the other three models. In summary, the CEEMDAN-VMD-BiLSTM rainfall prediction model proposed in this paper combines the advantages of various methods and has shown good predictive effects in experiments, providing an effective prediction method for rainfall.</p>","PeriodicalId":808,"journal":{"name":"Water, Air, & Soil Pollution","volume":null,"pages":null},"PeriodicalIF":3.8000,"publicationDate":"2024-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Rainfall Prediction Model Based on CEEMDAN-VMD-BiLSTM Network\",\"authors\":\"Sen Hou, Qikang Geng, Yaru Huang, Zhen Bian\",\"doi\":\"10.1007/s11270-024-07299-8\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<p>Rainfall prediction, based on meteorological data and models, forecasts the possible rainfall conditions for a period in the future. It is one of the important issues in meteorology and hydrology, and holds significant scientific and social value for enhancing human society's adaptive capacity, reducing the risk of natural disasters, promoting sustainable development, and protecting the environment. This study proposes a rainfall prediction model based on CEEMDAN-VMD-BiLSTM, which couples CEEMDAN (Complete Ensemble Empirical Mode Decomposition with Adaptive Noise), VMD (Variational Mode Decomposition), and BiLSTM (Bidirectional Long Short-Term Memory). The model first employs CEEMDAN and VMD, two decomposition algorithms, for a secondary decomposition of the original data, followed by prediction using the BiLSTM network. The study integrates the characteristics of CEEMDAN, which include adaptability, completeness, denoising capability, and high precision, the characteristic of VMD in extracting trend information, and the ability of the BiLSTM model to better capture contextual information in sequence data and solve long-term dependency issues, thereby increasing the accuracy of rainfall prediction. The research selected Zhongwei City in the Ningxia Hui Autonomous Region as the study object and used 20 years of monthly rainfall data from 2001 to 2020 as the research data. The model was compared with standalone BiLSTM models, CEEMDAN-BiLSTM coupled models, and VMD-BiLSTM coupled models. The model was validated using four indicators: RMSE, MARE, MAE, and NSE. The results showed that the maximum relative error of the CEEMDAN-VMD-BiLSTM neural network rainfall prediction coupled model was 7.22%, and the minimum relative error was -7.03%. The prediction qualification rate was 100%. The overall NSE value of the model ranged from 0.63 to 0.97, with most values between 0.86 and 0.97. The excellent rate was about 84.6%, and the good and above rate was 92.3%. In the rainfall prediction for Zhongwei City, the prediction accuracy of this coupled model was better than the other three models. In summary, the CEEMDAN-VMD-BiLSTM rainfall prediction model proposed in this paper combines the advantages of various methods and has shown good predictive effects in experiments, providing an effective prediction method for rainfall.</p>\",\"PeriodicalId\":808,\"journal\":{\"name\":\"Water, Air, & Soil Pollution\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":3.8000,\"publicationDate\":\"2024-07-02\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Water, Air, & Soil Pollution\",\"FirstCategoryId\":\"6\",\"ListUrlMain\":\"https://doi.org/10.1007/s11270-024-07299-8\",\"RegionNum\":4,\"RegionCategory\":\"环境科学与生态学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q2\",\"JCRName\":\"ENVIRONMENTAL SCIENCES\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Water, Air, & Soil Pollution","FirstCategoryId":"6","ListUrlMain":"https://doi.org/10.1007/s11270-024-07299-8","RegionNum":4,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"ENVIRONMENTAL SCIENCES","Score":null,"Total":0}
Rainfall Prediction Model Based on CEEMDAN-VMD-BiLSTM Network
Rainfall prediction, based on meteorological data and models, forecasts the possible rainfall conditions for a period in the future. It is one of the important issues in meteorology and hydrology, and holds significant scientific and social value for enhancing human society's adaptive capacity, reducing the risk of natural disasters, promoting sustainable development, and protecting the environment. This study proposes a rainfall prediction model based on CEEMDAN-VMD-BiLSTM, which couples CEEMDAN (Complete Ensemble Empirical Mode Decomposition with Adaptive Noise), VMD (Variational Mode Decomposition), and BiLSTM (Bidirectional Long Short-Term Memory). The model first employs CEEMDAN and VMD, two decomposition algorithms, for a secondary decomposition of the original data, followed by prediction using the BiLSTM network. The study integrates the characteristics of CEEMDAN, which include adaptability, completeness, denoising capability, and high precision, the characteristic of VMD in extracting trend information, and the ability of the BiLSTM model to better capture contextual information in sequence data and solve long-term dependency issues, thereby increasing the accuracy of rainfall prediction. The research selected Zhongwei City in the Ningxia Hui Autonomous Region as the study object and used 20 years of monthly rainfall data from 2001 to 2020 as the research data. The model was compared with standalone BiLSTM models, CEEMDAN-BiLSTM coupled models, and VMD-BiLSTM coupled models. The model was validated using four indicators: RMSE, MARE, MAE, and NSE. The results showed that the maximum relative error of the CEEMDAN-VMD-BiLSTM neural network rainfall prediction coupled model was 7.22%, and the minimum relative error was -7.03%. The prediction qualification rate was 100%. The overall NSE value of the model ranged from 0.63 to 0.97, with most values between 0.86 and 0.97. The excellent rate was about 84.6%, and the good and above rate was 92.3%. In the rainfall prediction for Zhongwei City, the prediction accuracy of this coupled model was better than the other three models. In summary, the CEEMDAN-VMD-BiLSTM rainfall prediction model proposed in this paper combines the advantages of various methods and has shown good predictive effects in experiments, providing an effective prediction method for rainfall.
期刊介绍:
Water, Air, & Soil Pollution is an international, interdisciplinary journal on all aspects of pollution and solutions to pollution in the biosphere. This includes chemical, physical and biological processes affecting flora, fauna, water, air and soil in relation to environmental pollution. Because of its scope, the subject areas are diverse and include all aspects of pollution sources, transport, deposition, accumulation, acid precipitation, atmospheric pollution, metals, aquatic pollution including marine pollution and ground water, waste water, pesticides, soil pollution, sewage, sediment pollution, forestry pollution, effects of pollutants on humans, vegetation, fish, aquatic species, micro-organisms, and animals, environmental and molecular toxicology applied to pollution research, biosensors, global and climate change, ecological implications of pollution and pollution models. Water, Air, & Soil Pollution also publishes manuscripts on novel methods used in the study of environmental pollutants, environmental toxicology, environmental biology, novel environmental engineering related to pollution, biodiversity as influenced by pollution, novel environmental biotechnology as applied to pollution (e.g. bioremediation), environmental modelling and biorestoration of polluted environments.
Articles should not be submitted that are of local interest only and do not advance international knowledge in environmental pollution and solutions to pollution. Articles that simply replicate known knowledge or techniques while researching a local pollution problem will normally be rejected without review. Submitted articles must have up-to-date references, employ the correct experimental replication and statistical analysis, where needed and contain a significant contribution to new knowledge. The publishing and editorial team sincerely appreciate your cooperation.
Water, Air, & Soil Pollution publishes research papers; review articles; mini-reviews; and book reviews.