Time series anomaly detection is critical in various domains, including stock markets, network traffic monitoring, and industrial systems, as it identifies deviations from expected patterns in data, enabling real-time analysis and timely responses to potential issues such as system failures or fraudulent activities. In the realm of industrial intelligent operation and maintenance, effective anomaly detection is essential for ensuring product quality and maintaining the safety of production systems. However, challenges such as large-scale data influx during testing and evolving data patterns, known as concept drift, pose significant challenges to the adaptability and accuracy of traditional deep learning methods. To address these challenges, we propose a novel plug-and-play Temporal Adaptation Anomaly Detection (TAAD) framework. This framework can seamlessly integrates with existing deep anomaly detection models, enhancing their capability to manage rapidly changing data patterns and concept drift. It comprises two key modules: a multi-scale prediction correction module that provides accurate predictions of future values based on historical data, and a memory comparison monitoring module that continuously updates memory vectors to detect segment anomalies and adapt to shifts in data distributions. Our TAAD framework substantially improves detection accuracy and efficiency, offering a robust and adaptable solution for real-time anomaly detection in industrial scenarios. We conducted experiments on the KPI datasets and demonstrated that our framework can effectively address concept drift.