Journal of Forecasting, Journal Year: 2024, Volume and Issue: unknown

Published: Nov. 18, 2024

ABSTRACT In light of the mounting environmental pressures, especially significant threat urban air pollution poses to public health, there arises an imperative need develop a data‐driven model for prediction. However, contemporary deep learning techniques, such as recurrent neural networks, often struggle effectively capture underlying data patterns and distributions, resulting in reduced stability. To address this gap, study introduces ensemble Wasserstein generative adversarial network framework (EWGF) enhance stability accuracy PM 2.5 predictions by facilitating acquisition more informative representations through network. The contains intricate feature extraction pipeline that automatically learns features containing residual information potential features, ameliorating underutilization information. We nonconvex multi‐objective optimization problem associated with amalgamating diverse architectures, which inherent instability predictions. Furthermore, adaptive search strategy is introduced ascertain optimal distribution prediction residuals, thereby expanding interval estimation method based on distribution. rigorously evaluate proposed using datasets from three major Indian cities, our experiments unequivocally show EWGF outperforms existing solutions both point prediction, evidenced approximate 8.07% reduction mean absolute percentage error 19.41% improvement score compared baseline model.

Language: Английский