LightGBM-driven one-day-ahead electrical load forecasting for enhancing operational efficiency of a power distribution company of Bangladesh

Short-term load forecasting (STLF) is a paramount progression in effective resource allocation, strategic infrastructure planning, and efficient operation of power systems. Accurate STLF ensures optimized power plant operations, grid stability, and energy conservation that positively impact the economic growth of a country. STLF is critical for any developing country like Bangladesh because accurate STLF can reduce drastically the power system operation and planning expenses. Dhaka Power Distribution Company (DPDC) Limited is a public limited company under the Power Division of the Ministry of Power, Energy and Mineral Resources of the Government of Bangladesh that manages the distribution of electricity to the customers in Dhaka City Corporation area. DPDC regularly provides one-day-ahead daily load consumption information to the National Load Dispatch Centre (NLDC), which is engaged in the management of critical functionalities of the power system network including electrical load demands across the country. DPDC mostly relies on assumption-based forecasting technique which often leads to significant errors in daily load demand estimation resulting in improper utilization of energy resources. This manuscript focuses on enhancing accuracy of one-day-ahead STLF for the DPDC by employing machine learning algorithms. Eight machine learning algorithms have been utilized in this study to assess the model accuracy and to identify the best-suited machine learning model for the one-day-ahead STLF. Among all the machine learning models, LightGBM model proved to be the most effective approach for STLF for DPDC dataset. Results suggest that LightGBM model exhibits average 5.08% reduction in MAPE, 47.76 MW decreases in RMSE, 5268.46 reductions in MSE, and 45.25 MW decreases in MAE from current assumption-based DPDC predicted method. These accuracy enhancements hold substantial economic benefits of Bangladesh aiding in the minimization of energy losses and optimization of power system operations. To validate the reliability of the used models, another dataset from the East Kentucky Power Cooperative (EKPC) Ltd. in the USA has been analysed by all the used ML models where LightGBM model exhibits the superior performance as observed in the case of DPDC. This study contributes to advancing the utilization of machine learning models in load forecasting techniques and thus offering valuable insights for improving efficiency and accuracy in power system management.