TUBITAK Project, 2023 - 2024
Renewable energy systems are becoming more important in our lives every day. Hybrid approaches are available for the efficiency and uninterrupted operation of renewable energy systems. However, there are also manufacturers that only use renewable energy systems and solar or wind farms are installed in a certain area. A plant that uses only a renewable energy system must use battery systems to adapt to instantly changing weather conditions or to prevent it from falling below the energy it needs to produce. The use of these systems results in additional costs. The aim of our study is to solve one of the most important problems faced by installations that have difficulty in predicting how much electrical energy they will produce on a daily basis for different companies, and to ensure that they meet this situation with minimum loss due to changing weather conditions, and in this way it is thought that our country will gain engineers who can make innovative projects by bringing a new perspective to renewable energy. Total daily solar radiation is considered the most important parameter in estimating the performance of renewable energy. The sizing of energy systems, especially photovoltaic systems, is of great importance in agricultural and building design applications. The method proposed by Hui-Min Zuo et al. Due to the natural variability of solar radiation, accurate short-term solar radiation forecasting is crucial for the stabilisation and cost-effective operation of power grids. Cloud cover is the most important factor influencing minute-scale variations in solar radiation. This study proposes a short-term solar radiation forecasting model based on a deep learning network. First, a novel hybrid cloud detection method is proposed to calculate the cloud coverage under different sky conditions. The relationship between the cloud coverage at the current time and the Global Horizontal Irradiance (GHI) 10 minutes later is directly analysed to avoid the error of assuming cloud motion as fixed motion. Meteorological parameters that have a strong correlation with the 10-minute GHI are used as model input in this study, mainly including relative humidity and 500 nm aerosol optical depth (AOD). The GHI of the next two time intervals are also selected after autocorrelation analysis to represent other large-scale effects. Together with the cloud, the meteorite parameters and the historical GHI are simultaneously used as input to a deep learning network based on a long short-term memory network (KSBA) and optimised by Bayesian Optimisation (BO). Solar irradiance ten minutes later is the output of the model. Experiments are conducted to evaluate the performance of the proposed prediction model by comparing it with some benchmark models. The results show that the proposed model outperforms the other models with a normalised root mean square error (nRMSE) of 15.25% under all-sky conditions. An improvement of 8.23% was achieved over the persistent model. The main energy sources of the project are solar energy. These sources are simulated in programmes such as Simulink and Python before the system is installed and circuit designs are made, and the results are predicted by the system. Deep learning algorithms will be used to control the system, use artificial intelligence algorithms and detect artificial solar power plants or artificially created roof panels. Then, how much solar radiation will be predicted by deep learning algorithms from sky images with deep learning algorithms and how much electrical energy the solar field or rooftop panel will produce in a short time will be estimated.