The rapid integration of renewable energy sources significantly complicates operational management and creates strong price fluctuations in the electric power and capacity market. Renewables, especially solar and wind power, are intermittent and have a stochastic nature, exacerbating the problem. This gives rise to the development of applied research in the field of renewable energy forecasting. One of the increasingly popular areas is the use of machine learning methods in forecasting problems in the electric power industry. In this paper, a detailed comparative analysis of two approaches to wind turbine output energy forecasting was carried out. The first one is to produce forecast of wind turbine output energy from wind speed forecast, while the second one is to directly forecast using historical data. Machine learning models were developed through these approaches accordingly and they were assessed using various performance evaluation metrics. The results indicate that forecasting errors of both methods' models have a linear relationship. It suggests the possibility of power forecasting error estimation through wind speed forecasting error. The last five-year weather data of the Krasnodar Territory coastal area were used. Experimental research was conducted on mathematical models of wind turbines that are installed in wind farms in the region. [ABSTRACT FROM AUTHOR]