This study is focusing on presenting an online machine learning algorithm that benefits from sequential data of IoT devices in the smart grid. This method provides the smart grid operator with the historical data of generation units of a smart grid that is connected to the IEEE 33-bus test system. The proposed smart grid consists of two photovoltaic cells, two wind turbines, a microturbine, a fuel cell and an electric car the behaviour of which is considered similar to that of a storage unit. In the training phase, the optimized generation units’ data is used to form a regressive model of every unit’s behaviour. Afterwards, the model is used to predict the behaviour of every unit in the next 24 h. The optimized operation data is used to solve the optimal power flow (OPF) problem. The output of OPF is useful in monitoring the stability of the smart grid, calculating power losses and locating possible faults. Moreover, the proposed framework benefits from the online discrepancy test (ODIT) method, which uses the data of the machine learning method to form a baseline for anomaly detection. The advantage of this method is that it minimizes false alarms and it eliminates false data in anomaly detection. The implementation of the proposed solution methodology has proven to be effective in regards with execution-time reduction and accuracy.