Power losses exist naturally and have to be cared for in the operation of electrical power systems. Many researchers have worked on various methods and approaches to reduce losses by incorporating distributed generators (DG), particularly from renewable sources. These studies are based on the maximum unit penetration of the DGs, which is rarely achieved, resulting in inaccurate calculations. This paper proposes an advanced solution for calculating power losses by incorporating an Extreme Learning Machine (ELM) method for forecasting the solar irradiation. The ELM algorithm was used to create a model for forecasting solar radiation in the Manokwari region and its surroundings. Daily solar radiation in the region has been predicted using the model. NASA's 8016 data on temperature and solar irradiation were used to train the ELM model. With an MAE value of around 0.6392 and a training time of 4.4375 seconds, the test results demonstrate that the built model has good accuracy. The operation of a 1000 kWp solar power plant based on the ELM data forecasting can reduce the power loss of the existing distribution network around the location from 1.5095 kW/hour to 0.9068 kW/hour. Furthermore, the power plant operation can minimize the power loss by 39.9249 percent, from 36.2280 kW to 21.7640 kW.