Smart meter data from Smart grid can be examined to identify anomalies in a number of areas, such as energy theft, cyber security, and defect detection. When it comes to anomaly detection, deep learning offers many benefits. From the unprocessed grid data, features must be extracted. Anomalies are defined as any events or modifications in the smart meter data that do not follow the usual pattern. The results of a typical grid configuration could vary substantially based on patterns or modifications on voltage, current, power, or consumption. A model for detecting anomalies for a hardware-based test bed implementation of an actual smart grid system is created in this study. By detecting unusual activity, the systemic actions in the flow of data transmission can be improved. It will additionally determine any modifications to parameters that may imply the existence of cyber attacks. Our suggested anomaly detection method makes use of federated learning and a Neural Network with Recurrent architecture (RNN) to distinguish outliers from regular observations. The effectiveness of the proposed detection technique was validated with a dataset collected from multiple Internet of Things devices, including smart meters, which served as a sample of the smart grid.