Smart Energy Management Systems (SEMS) employ Demand Response (DR), an effective vehicle for enabling utilities to manage resources efficiently, encouraging consumers to either cut or reschedule the demand at peak hours with appropriate market-based incentives. High accuracy of load forecasting methods (LF) is essential to make DR programs and resource management more effective. This research presents a novel approach to load forecasting with different multivariate datasets using 1D CNN, Bidirectional Long Short-Term Memory (Bi-LSTM), and Gated Recurrent Unit (GRU. Using Seasonal Trend Decomposition using LOESS (STL) it can detect more complex development trends and long-range dependencies, thus improving the forecast accuracy by combining load, price as well as weather data. The hybrid model yielded higher R2, RMSE, and MAPE scores compared to the current state-of-the-art algorithms. This study will be a key driver in the move towards developing more dynamic energy management policies for SEMS.