Procedia Computer Science · 2018 · 272 citations · 18 references
ComputingEnvironmental EngineeringEnvironmental SciencesAir PollutionForecastingMachine LearningAir PollutantsAir QualityDeep LearningParticulate MatterTime Series ForecastingTime Series PredictionAir Pollution ForecastingDeep Learning For ForecastingDeep Learning-based ForecastingGated Recurrent Unit NetworksLong Short-term Memory NetworksPollutant Concentration PredictionRecurrent Neural Network ForecastingRecurrent Neural Networks
Urbanization and industrialization have accelerated air pollution in developing countries, raising health and economic concerns, yet traditional prediction methods are computationally intensive and yield unsatisfactory results. The study applies deep learning models to improve air quality prediction. We trained RNN, LSTM, and GRU models on PM10 and meteorological time series from AirNet, exploring various architectures and hyperparameters over up to 1000 epochs with learning rates between 0.01 and 0.5. All three deep learning models achieved comparable prediction performance.
With the quick advancement of urbanization and industrialization, air pollution has become a serious issue in developing countries. Governments and natives have raised their increasing concern regarding air contamination since it influences human well-being and economic advancement around the world. Traditional air quality prediction methods depend on numerical data and require more computational power for the estimation of pollutant concentration and thus producing an unsatisfactory result. To tackle this problem, we applied widely used deep learning model. The pollutant considered for this work is Particulate Matter 10 (PM10). In this framework, Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU) are used for forecasting, based on the pollution and meteorological time series AirNet data. To figure out best architecture, we examined extensive analysis of different RNN models and its variations with its topologies and model parameters. Every experiment was run up to 1000 epochs by varying the learning rate in the range [0.01, 0.5]. It is observed from the study that all the three models performed comparatively well in prediction.
18
TensorFlow: A system for large-scale machine learning
Martı́n Abadi, Paul Barham, Jianmin Chen et al. · arXiv (Cornell University) · 2016 · 8.8K citations · Full text
Convolutional LSTM Network: A Machine Learning Approach for Precipitation Nowcasting
Xingjian Shi, Zhourong Chen, Hao Wang et al. · arXiv (Cornell University) · 2015 · 6.6K citations · Full text
TensorFlow: a system for large-scale machine learning
Martı́n Abadi, Paul Barham, Jianmin Chen et al. · Operating Systems Design and Implementation · 2016 · 6.4K citations
Backpropagation through time: what it does and how to do it
Paul J. Werbos · Proceedings of the IEEE · 1990 · 4.9K citations