Modern power systems’ day-to-day functions, such as economic operation, optimal control, and demand response, rely strongly on electricity demand characteristics. Notably, residential demand is rising faster and major contributing to demand response. Forecasting single residential consumer load is a difficile task due to consumption depending on the residents’ behavior, which is highly volatile. The article proposes an optimized long short-term memory (LSTM) network for short-term residential demand forecast to improve day-ahead forecast accuracy. For the current work, the benefits of gradient flow control and efficient management of long-term dependencies make the use of LSTM for forecast work. The hyperparameters of the LSTM network are optimally determined using an enhanced JAYA optimization algorithm. The proposed methodology is assessed on the EnergyPlus residential reference building consumption profile and (Market Analysis and Information System) MAISY residential load. The results have shown the proposed method is superior to traditional LSTM in terms of performance metrics.