Ανάλυση και πρόβλεψη της ατμοσφαιρικής μόλυνσης σε πολυάριθμες χώρες μέσω χρήσης BiLSTM-Conv1D νευρωνικών δικτύων
Analysis and prediction of air pollution in multiple countries using BiLSTM-Conv1D neural networks
Master Thesis
Author
Καράμπελας, Γεώργιος
Karampelas, Georgios
Date
2021-11View/ Open
Keywords
LSTM ; Conv1d ; Νευρωνικά δίκτυα ; Neural networks ; Μηχανική μάθηση ; Πρόβλεψη ; Prediction ; Ατμόσφαιρα ; Μόλυνση ; Air ; PollutionAbstract
Artificial Neural Networks is a scientific area, that has developed incrementally over the years and has been adopted to numerous fields of study due to its ability to process and discover patterns from raw data. The goal of a neural network model is to solve computational problems using different techniques and mathematical processes. One such problem is air pollution, an issue that is becoming more and more severe for the health of the general public. Governments and individuals require a way to know ahead of time and have an insight into how the quality of the air will be. That helps them take the initiative and act accordingly.
The goal of this thesis is to develop a neural network model that will take as inputs historical data of the atmosphere, and it will be able to predict the future values of the pollutants that affect the quality of air. The implementation will be accomplished with the programming language Python and the usage of libraries that were developed for machine learning.
In the context of the master thesis, 6 research papers were studied for the different types of neural network models on data related to air pollution and time series forecasting. Furthermore, multiple different types of neural networks were developed in order to realize the final architecture of the proposed model. In addition, plentiful experimentation was conducted to determine the best hyperparameters for the model and was assessed both for its performance and accuracy but also for its generalization to more than one location around the globe.