Using deep learning and an enhanced extreme learning machine based on genetic algorithms to forecast air quality index

Abstract

For as long as anybody can remember, air pollution has been a major social and public health issue. Predicting the AQI (Air Quality Index) using artificial intelligence algorithms allows for more comprehensive monitoring of future changes in air quality. Predicting air quality with one model of machine learning may be difficult when dealing with different patterns in AQI fluctuations. An enhanced extreme learning machine prediction approach based on genetic algorithms (GA-KELM) is developed to tackle this issue. To begin, a kernel approach is used to generate the kernel matrix, which subsequently takes the place of the hidden layer's output matrix. In order to fix the problem with traditional RLMSs, wherein the network's learning capacity is reduced due to the amount of concealed nodes and the randomly generated generation of weights and thresholds, a method called genetics is employed to optimize the kernel RLMS's hidden node and layer counts. Using the weights, root mean square error, and thresholds, the fitness function is defined. Lastly, the model's output weights are computed using the least squares approach. By iteratively optimizing the model's performance, genetic algorithms may discover the best option in the search space. To test GA-KELM's predictive capabilities, we use basic data from a Chinese city's long-term air quality forecasts to train a customized a kernel extreme learning equipment to forecast air quality variables including $SO_{2}$, $NO_{2}$, $PM_{10}$, $CO$, $O_{3}$, $PM_{2.5}$ concentration, and AQI. We compare this model to others, including Community Multiscale Air Quality, Support Vector Machines, and Deep Belief Connections with Back-Propagation. The findings demonstrate that the suggested model achieves better prediction accuracy and quicker training times.