A new cross-domain prediction model of air pollutant concentration based on secure federated learning and optimized LSTM neural network

As air pollution worsens, the fast prediction of air pollutant concentration becomes increasingly important for public health. This paper proposes a new cross-domain prediction model of air pollutant concentration based on federated learning (FL), differential privacy laplace mechanism (DPLA) and lo...

Celý popis

Uloženo v:

Podrobná bibliografie
Vydáno v:	Environmental science and pollution research international Ročník 30; číslo 2; s. 5103 - 5125
Hlavní autoři:	Huang, Guangqiu, Zhao, Xixuan, Lu, Qiuqin
Médium:	Journal Article
Jazyk:	angličtina
Vydáno:	Berlin/Heidelberg Springer Berlin Heidelberg 01.01.2023 Springer Nature B.V
Témata:	air air pollutants Air Pollutants - analysis Air pollution Air Pollution - analysis Algorithms Aquatic Pollution Atmospheric Protection/Air Quality Control/Air Pollution China data collection Domains Earth and Environmental Science Ecotoxicology Environment Environmental Chemistry Environmental Health Environmental Monitoring - methods Environmental protection Environmental science Federated learning Learning Machine learning Mathematical models memory Meteorological data Neural networks Neural Networks, Computer Outdoor air quality Parameters Passeriformes Pollutants Pollution abatement pollution control Pollution prevention prediction Prediction models Public health Research Article Rivers Search algorithms Short term memory Waste Water Technology Water Management Water Pollution Control China Sparrow search algorithm (SSA) Differential privacy laplace mechanism (DPLA) Air pollutant concentration prediction; Federated learning (FL) LSTM neural network Joint prevention and control
ISSN:	0944-1344, 1614-7499, 1614-7499
On-line přístup:	Získat plný text
Tagy:	Přidat tag Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!

Popis
Shrnutí:	As air pollution worsens, the fast prediction of air pollutant concentration becomes increasingly important for public health. This paper proposes a new cross-domain prediction model of air pollutant concentration based on federated learning (FL), differential privacy laplace mechanism (DPLA) and long and short-term memory network optimized by sparrow search algorithm (SSA-LSTM), named FL-DPLA-SSA-LSTM. Firstly, with FL, SSA-LSTM is used as local training model for each city and predicts air pollutant concentration. Secondly, DPLA is used to add noise to the local model parameters, which can protect local data security. Then, the global model is updated by using the federated averaging algorithm (FedAvg). Lastly, FL is used to share global model for all cities, which can safely and quickly cross-domain predict air pollutant concentration. For data set, it is taken from hourly air pollutants and meteorological data from 12 cities in the Fenhe River and Weihe River Plains in China in 2020. The experimental results show that the prediction performance of the proposed model is significantly better than all comparison models. FedAvg updating with local model parameters with DPLA noise has little effect on the performance of the global model and even exceeds that of the global model. The calculation time of FL-DPLA-SSA-LSTM model is reduced by 99.95% compared with that of not using FL-DPLA machine learning model. It is proved that the model is high sharing and high safety, which greatly improves the training efficiency and has better generalization ability. It is significant for joint air pollution prevention and control and environmental protection.
Bibliografie:	ObjectType-Article-1 SourceType-Scholarly Journals-1 ObjectType-Feature-2 content type line 14 content type line 23
ISSN:	0944-1344 1614-7499 1614-7499
DOI:	10.1007/s11356-022-22454-6