Power generation of photovoltaic system depends on variability of weather and climate variables. For this reason, predictive models are required for forecasting power generation in order to improve performance of operations on the grid. The paper aims to develop single-step predictive models of output power with Machine Learning and Deep Learning techniques from an autonomous photovoltaic system that were trained with historical data acquired from solar irradiance, temperature, voltage, and current sensors with steps of 5 min that are stored in a database. In addition, the same procedure is applied for comparison of performance with historical values from a meteorological database stored each hour. Time-series models were developed using techniques like Random Forest (RF), Support Vector Machines for regression (SVR), Classical Neural Network (ANN), Recurrent Neural Network (RNN), Long Short-Term Memory Neural Networks (LSTM), and Convolutional Neural Network (CNN). Additionally, in order to compare the algorithms, some metrics were calculated. Those metrics were: mean absolute error, mean squared error and R2 coefficient. The results showed that LST and CNN Neural Networks had the lowest values of the error metrics and the highest values in R2 coefficient.