Sudha R, Damodaran Ajith, Manohar Gunaselvi
Department of Electronics and Instrumentation Engineering, Easwari Engineering College, Ramapuram, Chennai, 600089, Tamil Nadu, India.
Department of Mechanical Engineering, Saveetha School of Engineering, Saveetha Institute of Medical and Technical Sciences, Thandalam, TamilNadu, 602105, India.
Sci Rep. 2025 Aug 19;15(1):30428. doi: 10.1038/s41598-025-14668-8.
In most industrial and urban regions, monitoring and safeguarding the air's purity is considered one of the most crucial tasks for government agencies. In numerous industrial and urban locations, preserving and tracking the condition of the air has become the primary concern. However, implementing comprehensive air quality monitoring systems often requires significant financial investment. The performance of current air quality monitoring sensors declines over time, which leads to inaccurate measurements of air pollution levels. To address this challenge, it is essential to develop and implement strategies aimed at maintaining sensor accuracy and effectively resolving environmental issues related to air quality. To facilitate an effective air quality prediction and assessment, a deep learning network is proposed. At first, the data for predicting air quality are collected from the relevant data sources. The proposed model introduces a novel methodology for weighted feature selection utilizing an Improved Gannet Optimization Algorithm (IGOA) aimed at enhancing the performance of data classification. After extracting the weighted features, classification is carried out using an Adaptive Residual Bi-LSTM network combined with Pyramid Dilation (ARBi-LSTM-PD), which significantly increase the model's potential to identify complex patterns within the data. The efficacy of the implemented model is enhanced by optimizing the parameters from RBi-LSTM using the IGOA strategy. This approach tackles the difficulties associated with feature selection and classification, leading to distinct advancements in the quality of the classification results. The robustness of the model is examined and analyzed using different measures. The accuracy and precision rate of the proposed model are 95.175% and 87.2%, which is better than traditional air quality prediction models. Thus, the simulation results demonstrate that it obtains the desired results for predicting and assessing the quality of air.
在大多数工业和城市地区，监测和维护空气纯净度被视为政府机构最重要的任务之一。在众多工业和城市地点，保护和跟踪空气质量状况已成为首要关注点。然而，实施全面的空气质量监测系统通常需要大量资金投入。当前空气质量监测传感器的性能会随着时间下降，这导致空气污染水平的测量不准确。为应对这一挑战，制定并实施旨在维持传感器准确性以及有效解决与空气质量相关环境问题的策略至关重要。为促进有效的空气质量预测和评估，提出了一种深度学习网络。首先，从相关数据源收集用于预测空气质量的数据。所提出的模型引入了一种新颖的加权特征选择方法，利用改进的塘鹅优化算法（IGOA）来提高数据分类性能。提取加权特征后，使用结合金字塔扩张的自适应残差双向长短期记忆网络（ARBi-LSTM-PD）进行分类，这显著提高了模型识别数据中复杂模式的能力。通过使用IGOA策略优化RBi-LSTM的参数，增强了所实施模型的有效性。这种方法解决了与特征选择和分类相关的难题，使分类结果质量有明显提升。使用不同方法对模型的稳健性进行了检验和分析。所提出模型的准确率和精确率分别为95.175%和87.2%，优于传统的空气质量预测模型。因此，仿真结果表明它在预测和评估空气质量方面取得了理想的结果。
