Environmental pollution, particularly air pollution, has long been an issue in every major city on the planet. For many years, accurate estimation of PM2.5 and PM10[8] [12] fine dust concentration values has been a fascinating study area. This study focused on 3-Day Transboundary Air Pollution Prediction, which proposed to merge many deep learning models and pick appropriate properties for the PM10 index prediction problem by utilizing various features such as timestamps, geographical features, and public weather data. Using the dataset provided by MediaEval, we examined the performance of several learning models and features in order to investigate the problem. Experimental results show that combining multiple deep learning models together gives a higher overall performance than other techniques and features in RMSE, MAE, SMAPE.
Haze air pollution is defined as the presence in the air of particulate
matter such as smoke, dust, and other vapours that arise from the
large-scale forest and land fires, factories, and automobiles. When
the concentration of airborne pollutants reaches dangerous levels,
it causes respiratory issues and has significant consequences for
visibility, economic productivity, transportation, and tourism.[
Transboundary haze is a recurring problem in many parts of the
world, particularly in Southeast Asia, where haze pollution sources
difer from nation to country, with varying percentages coming
from localized or transboundary sources.[
The goal of this article is to address the sub-task 2 of the
competition which is to examine the transnational PM10 estimate problem
using timestamp information, location data, and weather data using
the technique of mixing multiple deep learning models.[
We’ll start with a high-level overview of the topic and then go over to our specific approach that we have developed for a PM10 prediction algorithm.
At this stage the data for each station is separated from the original data (train air quality and train weather) based on the station’s ID. Then we will combine the data by province as the data on air quality and weather can be found in the same province.
We’ll utilize the interpolation method[
With testing data (from 2018-2019), we’ll utilize the zero-fill method to generate a mask to help the model work with missing values of data that is used for prediction. 2.2
We model the PM10 index prediction problem as a regression problem with the following features to estimate the PM10 index in the near future from a list of specified attributes.
2.2.1 Timestamp features. Because outdoor air quality varies greatly depending on the time of day, timestamp information can be valuable for PM10 estimation dificulties. In particular, each country’s PM10 index is unique, as are the provinces within the same country. As a result, we created a PM10 index for a country by averaging the PM10 scores of the provinces. We take the daily average and then average it across provinces in provinces where PM10 is reported hourly.
2.2.2 Location features. We choose one province as a landmark, and then we evaluate the closest distances between provinces in the vicinity of the landmark as it can provide useful information for plot analysis and reduce noise due to air pollution in the data.
We calculated the distance using the Haversine formula[
2.2.3 Weather features. Public weather features include information on weather, such as "temperature," "precipitation," "humidity," "wind direction," and "wind speed," obtained from local stations. These characteristics can be thought of as supplementary data that can help machine learning models become more robust and reliable.
In order to provide the best forecast results, we analyze the correlation between these weather features and the PM10 index in each country and select the features with the strongest correlation. The wind direction had a strong correlation with PM10 in the Indonesia dataset; the temperature, rain, and humidity in the Brunei dataset; the rain, humidity, and wind speed in the Thailand dataset; and the temperature in the Singapore dataset.
If some of the provinces are lacking "Temperature" data We
utilize a basic LSTM[
To train a model that is appropriate for the problem, we split the raw training dataset (2010-2017) into training and test datasets, using 80% of them for training and the remaining 20% for model testing. After that, we will predict PM10 on testing dataset (2018-2019). This study uses a regression model to forecast PM10 for the next three days. 2.4
Process the data for each station in section 2.1, then build a dataset for each country by averaging the weather and air quality features of the stations in that country; for example, data of Brunei would be obtained by averaging the weather and air quality features of the stations in Brunei.
For each country, we use data from the previous three days to forecast the future three days.
We employ the concept of merging diferent models[
We’ll need three branches to construct our multi-input network:
The first two forks will be a simple BiLSTM[
We must fill in the mean from 2010 to 2015 because Singapore weather data is only available from 2016 to 2017, hence we do not recommend using Singapore data to train the multimodal model to avoid overfitting the model. We’ll run the BiLSTM model for Singapore data separately. 2.5
We employ root mean square error (RMSE), mean absolute
error (MAE)[
The organizer has provided us with a data set. Weather and air quality data are included in the dataset, with the training set spanning 2010-2017 and the test set spanning 2018-2019.
T.Nguyen 3.2
To improve the performance of the presented approaches, we use a random search method to select ideal hyperparameters based on performance indicators. For each model, this method entails scanning a predefined parameter space and picking the best performing hyperparameters which is shown in Table 1. 3.3
In this study, we compare the proposed methodologies’ performance to the performance measures listed in Section 2.5.
The MAE and RMSE results of multimodal are better than the rest of the models, as shown in Table 2. However, because multimodal employs the average PM10 value of countries, the forecast results will be skewed for countries with excessively low or excessively high PM10.
The MAE and RMSE values in Table 2 are better than those in Table 3, but the SMAPE results are worse. 4
After three months for analyzing, we illustrated the benefits of combining numerous models with generalization and deep learning methods to address the PM10 index estimation problem utilizing various types of features such as timestamps, location, and public weather data. By incorporating a variety of characteristics, the test results reveal that PM10 level prediction is fairly accurate when compared to ground-truth. Transnational air pollution can be predicted using the strategy of merging multiple models.
We’re excited to continue our research by examining additional forms of data, such as image data, video, and new deep learning models. One of the new approachs is multivariate transformer learning because this learning can learn across long timespans.