A Deep Hybrid Learning Model for Photovoltaic Solar Tracking Systems

Abstract

The demand for clean and sustainable energy has increased the popularity of photovoltaic (PV) solar panels. PV solar panels are the most effective technology for transforming the sun's rays into a valuable energy source. To maximize the amount of energy captured, the rays of light from the sun must be at a 90-degree angle to the surface of the PV panel. To accomplish this, PV panels are modified with solar trackers that can effectively track the sun's position as it shifts during the day. Due to the randomness and nonlinearity of metrological data, using deep learning (DL) algorithms to enhance solar trackers has gained much popularity among researchers. However, given the varying nature of metrological data, single DL models sometimes fail to perform satisfactorily. For this purpose, this study applies sine and cosine transformations (SCT), a convolutional neural network (CNN), and long short-term memory (LSTM) to forecast the sun's trajectory. The SCT captures the cyclic components of the metrological data. The transformed data is then fed into a CNN-LSTM framework, where features of the sun’s movement are learned. The SCT-CNN-LSTM framework’s performance is evaluated concerning three other models based on the MAE, MAPE, and RMSE performance metrics. Based on the research findings, the CNN-LSTM framework has a superior performance, achieving an MAE, MAPE, and RMSE score of 0.0010, 73.5%, and 0.0012, respectively.