The increasing concern for alternative energy sources presents a requirement for the identification of innovative methods to expand the capacity of solar-generated energy. This research contributes to research on a new deep learning-based solar tracking system. It uses long short-term memory (LSTM) model with a Convolutional Neural Network (CNN) to maximize the efficiency and performance of solar panels and predict real-time optimal tilting or rotation angles to achieve more efficient energy performance. The system uses an IoT-enabled framework that depends on sensor readings such as light intensity and environmental conditions for the tracking system. The results from the experiment show that CNN with LSTM outperformed traditional tracking methods and deep learning architectures, where the Mean Absolute Error (MSE) was 0.15, the Mean Absolute Percentage Error was 4.5%, and the Root Mean Square Error (RMSE) was 0.20. The results obtained indicate that large amounts of electricity are being produced since the peak values are found to be 8.0 kWh when the model involved the CNN with LSTM, whereas only 4.5 kWh was produced when the latter model was not involved. At last, the precision, recall, F1 score, and accuracy of the model during the use of CNN with LSTM are at 92.0%, 90.0%, 91.0%, and 95.5%, respectively; this indicates the models had a high probability of obtaining maximum solar energy capture. It reveals 460 true positives and only 20 false positives, thus confirming the reliability of the model. The real-time implementation of the proposed model highly improves energy output while offering an adaptive solution for solar energy systems in modern grid infrastructures. Also, this research emphasizes the need to integrate modern deep learning approaches into a solar tracking system. This helps improve efficiency in the generation of sustainable clean energy that adheres to global initiatives of cleaner solutions.