The state-of-the-art causal discovery algorithms are typically based on complete observed data. However, in reality, technical issues, human errors, and data collection methods among other reasons result in missing data. The methods for handling missing data mainly involve statistical and machine learning approaches, where statistical methods are simple and practical, while machine learning methods offer higher accuracy. The typical approach for causal structure discovery in the presence of missing data involves two steps: First, applying missing data imputation algorithms to address the issue of missing data, and then using causal discovery algorithms to identify the causal structure. However, this two-step approach is suboptimal because imputing missing data may introduce biases in the underlying data distribution, making it challenging to accurately assess causal effects between variables. This paper proposes an iterative approach based on generative models for both missing data imputation and causal structure discovery. This approach incorporates an architecture based on Wasserstein generative adversarial networks and autoencoders (AE) to respectively impute missing data and output the causal structure. Through extensive experiments comparing against various state-of-the-art baseline algorithms, the effectiveness and superiority of this method are validated, providing valuable insights for further research on causal structures in the context of missing data.