Recently, the share of renewable energy sources in the structure of production units of the energy sector has been steadily increasing. The unpredictability of renewable energy sources leads to difficulties in managing, planning and development of the power system. One of the ways to reduce the negative impact of unpredictable renewable energy sources is to predict the availability of these sources. Short-term forecasting of photovoltaic power plant production is one of the tools that make this possible. The information for the short-term production forecast model can be collected by taking successive photographs of the sky hemisphere above the photovoltaic power plant. By processing the data contained in a series of photographs, a trend can be identified that allows the prediction of the future production of the photovoltaic power plant. In this work, a new method for short-term production forecasting was explored that involves continuous photography of the sky above the photovoltaic power plant. In addition to the new method, a system for collecting and processing photographs of the sky hemisphere was developed to reduce the cost of integrating such a system. This system is not only cost-effective, but also enables the collection of photos with a high sampling frequency, which is necessary to capture information about rapid cloud changes over the photovoltaic power plant. A hybrid model integrating convolutional neural networks and long short-term memory was developed and tested for short-term production forecasting. This model effectively reveals spatial and temporal patterns from images and output power data, demonstrating significant prediction accuracy. The results of the developed model were compared with a threedimensional convolutional neural network model trained on the same data. This work highlights the potential of Deep Learning in driving innovation in renewable energy sector, particularly in power forecasting, contributing to the overall sustainability of electrical power systems.