关键词: 生成对抗插补网络, 缺失数据, 自注意力机制, 残差连接

Abstract: Water quality monitoring, air quality monitoring, and soil monitoring are crucial components of environmental monitoring. However, during actual operations, monitoring equipment may encounter various interferences such as equipment failures, sensor malfunctions, and data transmission issues, leading to missing monitoring data. Missing data can severely impact the assessment and analysis of environmental conditions. To address these issues, a missing data detection method based on the GAIN-ARN (Generative Adversarial Imputation Network with Attention and Residual Network) model is proposed. This method incorporates self-attention mechanisms and residual connections to enhance the model's ability to capture the internal structure of the data, thereby improving the stability and effectiveness of data imputation. Experiments conducted on five public datasets demonstrate that GAIN-ARN achieves the lowest RMSE values in almost all cases, with the most significant improvement observed in the Water Quality Testing dataset. At a 20% missing rate, GAIN-ARN shows an improvement of approximately 10.61% compared to GAIN and about 33.22% compared to WSGAIN-GP. The experimental results further indicate that the introduction of the self-attention layer significantly enhances the imputation performance of the generator. 

Key words: generative adversarial imputation network, missing data, self-attention mechanism, residual connection