The use of deep learning techniques in detecting anomalies in time series data has been an active area of research with a long history of development and a variety of approaches. In particular, reconstruction-based unsupervised anomaly detection methods have gained popularity due to their intuitive assumptions and low computational requirements. However, these methods are often susceptible to outliers and do not effectively model anomalies, leading to suboptimal results. This paper presents a novel approach for unsupervised anomaly detection, called the Cooperative Network Time Series (CNTS) approach. The CNTS system consists of two components: a detector and a reconstructor. The detector is responsible for directly detecting anomalies, while the reconstructor provides reconstruction information to the detector and updates its learning based on anomalous information received from the detector. The central aspect of CNTS is a multi-objective optimization problem, which is solved through a cooperative solution strategy. Experiments on three real-world datasets demonstrate the state-of-the-art performance of CNTS and confirm the cooperative effectiveness of the detector and reconstructor. The source code for this study is publicly available on GitHub.