The Water Distribution System (WDS) is one of the most critical infrastructures in modern society, which makes WDS become more vulnerable to the threat of cyber attack. Therefore, timely detection of cyber attack is crucial for WDS. However, existing attack detection methods face challenges in capturing both the complex temporal relationships and long-term dependencies of time-series data generated by WDS. To address these issues, we propose a detection model based on Adaptive Interactive Aggregation Convolution (AIAC) and REVIN normalization with Stabilized Long Short-Term Memory (RSLSTM). The AIAC module is employed to capture intricate temporal correlations across WDS components through the interactive aggregation of different convolutional layers with adaptive weights and channel attention mechanism, while the RSLSTM model is employed to better capture long-term dependencies of the data. We evaluated the proposed method on two WDS datasets. The experimental results demonstrate that our method exhibits robust performance and effective cyber attack detection capabilities.