<p>Recommendation algorithms based on graph convolutional networks can integrate user and item node information along with their interaction topology, better capturing the intricate relationships between users and items, thereby enhancing the accuracy of recommender systems. However, existing methods often overlook the impact of noise in user behavior data on recommendation performance. Additionally, when there are too many convolutional layers in the graph, the node representations tend to smoothing, resulting in an inability to accurately distinguish user preferences. To address these issues, we propose a self-supervised graph convolutional model for recommendation with exponential moving average (SGCERec). Specifically, we first employ exponential moving average (EMA) techniques from the field of time-series analysis to denoise the raw user interaction data. Then, by applying layer filtering technique to update the propagation of information and the representation of nodes within the graph convolutional network, we effectively deepen the model hierarchy, enabling the model to gain a deeper understanding of the features and structures of the graph data, thereby improving the performance and effectiveness of the recommender systems. Finally, experimental results on three real datasets show that SGCERec outperforms state-of-the-art recommendation methods across various common evaluation metrics.</p>