Abstract
The demand for electrical energy continues to rise with the progression of time. This growth must be matched by a reliable and cost-effective supply of electricity, requiring power systems that are both dependable and economical. Since the amount of electricity consumed by users cannot be precisely predicted, balancing generation with consumption necessitates accurate electrical load forecasting. This study focuses on load forecasting using the Adaptive Neuro-Fuzzy Inference System (ANFIS) method. The forecast developed targets daily peak loads, which fall under short-term load forecasting. The data used for this forecasting consists of historical daily peak loads from January 1, 2017, to June 9, 2022. The forecasting process involves parameters such as radius, squash factor, accept ratio, reject ratio, and epoch. The forecast accuracy is evaluated using the Mean Absolute Percentage Error (MAPE) metric. The results are then compared with PLN’s load forecasting, which employs the load coefficient method. The ANFIS-based forecasting achieved a MAPE of 1.879%, using networks Jaringan_24 and Jaringan_25. This MAPE value is slightly lower than PLN’s load forecasting MAPE of 1.917%, indicating better accuracy by the ANFIS method.