The electricity demand-supply imbalance in numerous developing countries results in power outages, creating a need for innovative solutions. We categorize power outages into scheduled and unscheduled power outages. While extensive research exists on Demand-Side Management (DSM) in stable power grids, applying DSM strategies in unreliable power grids remains largely unexplored. Consequently, our study aims to implement DSM, combining AI planning and Machine Learning, to mitigate the impact of planned and unplanned power interruptions in smart homes within unstable power networks. Therefore, this research introduces a hybrid approach merging knowledge-driven based Temporal Planning with data-driven based LSTM model, utilizing the former to address DSM strategy together with scheduled outages, and the latter to proactively handle unscheduled outages. Furthermore, our study generally extends AI planning and especially the Temporal Planning by applying Timed Initial Fluents (TIFs), which have seen limited practical implementation to date. TIFs allow the declaration of numeric fluents at specific time points, expanding the expressive power for addressing complex problems. The implemented solution involves applying LSTM networks on household power dataset and weather information to predict unscheduled power interruptions. Leveraging the learned predicted model and Temporal AI planner, the system creates plans to ensure uninterrupted power supply and minimize energy costs using DSM strategies. The electricity cost reduction is performed by shifting the loads of the smart home to the battery during peak tariff hours, thus implementing DSM based-TOU technique. The cost of battery charging is optimized by charging it at the off-peak hours. Hence, the results obtained in this thesis indicate that our hybrid approach is effective for implementing DSM in unreliable power grids.