Development of Liquid Neural Network (lnn) for Short Term Load Forecasting (stlf)

ABSTRACT In today’s rapidly evolving world, the demand for electrical power is increasing at an unprecedented rate, making its role in everyday life undeniable. The importance of a reliable power supply cannot be overstated, particularly as societies depend on continuous and stable access to electricity. However, rising demand presents the challenge of maintaining consistent supply amid fluctuating consumption patterns and weather-driven variations in load. This study explores the application of Liquid Neural Networks (LNNs) for short-term load forecasting (STLF), given their ability to capture dynamic, time-dependent fluctuations in energy consumption. Three LNN models were developed and tested: the first, trained on the PJME dataset using only historical load data, achieved a test mean absolute error (MAE) of 0.0387; the second, incorporating temperature as an exogenous factor, improved performance with a test MAE of 0.0361, validating the role of external variables; and the third, trained on synthetically generated data that mimicked Nigeria’s irregular load patterns, yielded a test MAE of 0.0734, reflecting the complexity of forecasting under highly volatile conditions. The findings confirm that LNNs are competitive forecasting tools with strong adaptability to dynamic data environments, while integrating exogenous factors enhances accuracy and the synthetic experiment demonstrates both the challenges and opportunities of applying LNNs to Nigeria’s power system. These results suggest that LNNs hold significant potential for improving grid reliability, operational planning, and energy management in emerging economies.