An Artificial Neural Network (ANN): Based air quality index forecast for the city of Meerut

Introduction

As industrialization advances, air pollutants such as particulate matter (PM10), carbon dioxide (CO2), and sulfur dioxide (SO2) are on the rise. This surge in population has a detrimental effect on the environment and human well-being, which adversely affects public economies. By implementing these measures, these air pollutants are effectively regulated, and state-run administrations, both local and general, respond appropriately. These toxins are the ones that are deemed significant when evaluating their impact on the environment and human wellbeing. One of the most harmful atmospheric pollutants to human health is SO2, as it primarily affects the respiratory system (Boznar et al., 1993(Boznar et al., & 2002) ) [3, 4] . It could potentially pose a risk to human health as it can be converted into sulfate and sulfuric acid. Both natural and artificial sources contribute to the production of SO2. The foundation of the essential natural resource is composed of volcanoes. One of the major causes that humans contribute to is the combustion of fuel, particularly diesel and coal. The majority of SO2 is generated by power plants, companies involved in metal processing and handling, and vehicles that rely on these fuels. These indicators are regularly evaluated and carefully scrutinized both within the country and on a global scale. The UPPCB's air observation facilities in Meerut keep track of the pollutants present in the air. These stations keep track of substances and gases that can impact the air's quality. The results are frequently updated on the website of the center. The purpose of this review is to assess the air quality in Meerut, the industrial hub of Uttar Pradesh, and to predict and visualize the situation using a multilayer perceptron (MLP) approach. The data collected by the UPPCB from 2019 to 2023 was analyzed in this context. Following that, the projections were obtained, and SPSS software was utilized to present the data between 2019 and 2023. In this process, the MLP approach was employed.

Methods

An artificial Neural Network (ANN) is a quantitative and computational model that draws inspiration from the structure and operations of real brain networks. They are mainly distinguished by their neural activation technique, preparation approach, network architecture, relationship example, and data handling capabilities. The NN model that is most frequently used is the MLP. This kind of neural network is referred to as a controlled network as it requires an ideal output in order to train. The goal of this kind of network design is to use verifiable data to build a model that accurately connects input and output. Therefore, the model can provide the intended outcome even if it is not immediately apparent. When data is moved from the input layer to the hidden one, the connection weights cause the number of data sources to rise. They are handled by a nonlinear function in the hidden layer once they have been summarized. The next hidden layer adds, enlarges, and controls the data that is handled by the connection weights after it exits the previous hidden layer if there are many hidden layers. The data is then copied with accessible weights and processed once again using the output layer to supply the neural network's output. Before the neural network can be applied to any specific task, it must be trained on several input-output mapping tests. For every trained neural network to produce reliable results, these are the necessary details. The sample utilized for training information must thus be rather large and comprise a large amount of data from various process variables and experimental conditions to contain all the pertinent information.

Time series interpretation

Multi-layer perceptron (MLP) is applied to finish the analysis of time series, and it is defined by:

Where 𝑇 1 , 𝑇 2 , 𝑇 3 , …, 𝑇 𝑞 are inputs and 𝑇 𝑞+1 is output. MLP is applied to form 𝑓. The output is defined as the fifth information point, following the use of the first four as input.

Data analysis

The SO2, PM10, and NO2 air pollution components are the three categories on which the UPPCB screens information for its webpage to compile information about the state's numerous metropolitan districts. The AQI levels are measured at Meerut City's Kesharganj Street. The stage has proven crucial during the preceding years. The quality of the air in Meerut would deteriorate in tandem with an increase in the AQI. In this work, AQI data for Meerut's Kesharganj Road has been gathered between January 2019 and June 2023. An overview of the data is shown in Figure 1:

Examination of ANN

Using SPSS programming, we have developed an MLP system for the town of Meerut. This inquiry uses 51 points of information in total. These noteworthy data points are categorized into 47 groups. For every group, there are five points of information. The result was considered to be the fifth AQI information point, with the starting four points serving just as input. Again, when the first information point is eliminated, the following four points of information are inputs, and the subsequent information point within the order will be output, and so on. The four underlying data inputs are A_A, A_B, A_C, and A_D, whereas the label assigned for the output data point is a Output. Below is a time series structure that has been developed. Network information, comprising input, hidden, and output layers, is displayed in Table 1. Relative error during testing is 0.425 and during training is 0.455, based on the model summary displayed in Table 2. Table 3 displays parameter estimation, while Table 4 discusses the significance of the independent variables. The specifics of the network's configuration are illustrated in Figure 2. In Figure 3, predicted values are displayed versus output. Figure 4 displays information about expected values in comparison to residuals. The normalized importance of the network's independent variables is displayed in Figure 5.

Conclusion

In this paper, time series analysis is used to determine the AQI levels. There is non-linearity in the model. To manage non-linearity, an ANN-based model is built. The MLP is employed when building the model using time series. According to the model, the relative error was 0.425 during the testing phase and 0.455 during the instruction period. This implies that future AQI values may be predicted using historical data.

References

1. Asadollahfardi G, Zangooei H, Aria SH. Predicting PM2.5 concentrations using artificial neural networks and Markov chain: A case study Keraj City. Asian J Atmos Environ. 2016;10(2):67-79.
2. Bhavsar R. Air pollution monitoring using artificial neural network. Int J Sci Eng Res. 2019;10(12):515-519.