Electrical loading prediction is a key aspect of the power system governing, operating, and scheduling. Energy suppliers can control the running system cost by using a lot of information it provides thereby optimizing the power system operation performance. The demand for the electricity well forcasted means more than half of their energy efficiency. Implementation of this work traces out an in-depth detail of integrated quality time series forecasting models on the prediction of electrical consumption. The primary goal of the study is to assess the performance of two state-of-the-art forecasting models: Deep LSTM version and long short-term memory (LSTM) neural networks, Seasonal autoregressive integrated ma. The main task is to evaluate the models’ precision in predicting daily energy consumption based on the historical demand data, holiday data and other time-related lines of evidence. The performance of the models is assessed based on the Mean Absolute Percentage Error (MAPE). The method covers feature engineering, the data preparation, model selection, and assessment. The generated MAPE values illuminated the performance of the models— SARIMA performed relatively inaccurately, and LSTM and deep LSTM significantly improved, obtaining a very good MAPEs of 7.5% and 7.45%, respectively. Notably, the deep LSTM version shows a superiority in prediction compared to other models, with particular emphasis on capturing the temporal relationships. This study makes a great contribution to the field of energy forecasting as it shows applicability of LSTM- and SARIMA- based models for the very good forecast of the consumption power. It captures the attention on how the LSTM networks at 20% of depth; may help in improving prediction accuracy when there are complex patterns and long-distance dependence is a concern. To utility companies, the grid operators and lawmakers who are out to harness every energy resource, to cut the costs, and ensure a continuous flow of electricity; such results are so very helpful.