<p>Advancements in technologies and lifestyles have greatly altered energy consumption patterns. This makes it more crucial to have accurate long-term load forecasting (LTLF) for the power system to function effectively. Mostly, LTLF has been carried out using an aggregate load at grid, substation, feeder or individual consumer’s level. Therefore, a persistent gap still occurs to carry out a novel LTLF based on load factor and real-time smart energy meters datasets for diverse consumers using hybrid convolution neural network and long short-term memory sequential modelling on monthly basis for three years. Moreover, a novel monthly average load analysis based on load factor is also carried out to capture the diversity of such consumers. This paper designs a novel technique for merging datasets of similar consumers to create dynamic data which involves time alignment, preserve all timestamps, handling missing values, compute sum of (active load, reactive power and currents) and average of (voltage and power factor), aggregate smart meter data across same timestamp and assemble clean data set. This dynamic data is used as input data for the LTLF phase and monthly average load analysis. The performance of the proposed model is validated and tested which outperformed the other baseline models in terms of overall average mean absolute percentage error (MAPE) across 12 folds i.e., 0.39, 1.46, 2.82, 3.24, and 3.27% for tubewells, hospitals, schools, offices, and colleges respectively, and confirms its exceptional accuracy with squared-R (R²⁾ value of 0.99 for (offices and hospitals), 0.98 for (tubewells and schools), and 0.95 for colleges. The achieved results also confirm that the consumer with a high load factor have less MAPE as compared to consumer having a low load factor.</p>