Monitoring of geological reservoirs using 4D seismic faces many challenges. The repeatability between different surveys needs to be optimal in which changes are only present in the target zone. Ideal conditions require having the same acquisition parameters for each survey and no near-surface variations, like those caused by seasonal changes. In practice, data processing and matching techniques are required to improve the repeatability of the data. This study proposes a deep learning approach for post-stack trace-by-trace matching to reduce the remaining 4D noise. We utilize the sequential nature of seismic data to train a temporal convolutional network (TCN), which learns to map the monitor traces to the base data in the overburden region. The goal is to suppress 4D noise while maintaining time-lapse signal caused by the reservoir changes we wish to monitor. We validate the method on synthetic time-lapse zero-offset data and show improvements in repeatability. We also perform an initial test on 4D land data to show the potential for application to real datasets.