Solar energy is used as electrical energy for daily life by using solar cell. There are several models for maximizing the absorption of solar cell power by using reflectors and solar tracking. Reflector focuses sunlight on the solar cell to get maximum light. Solar tracking real-time tracking the sun by moving two motors and sensors to read the direction of the sun's angle. Both have the disadvantage of tracking the sun continuously. So it is necessary to optimize the system to determine the tilt angle of the actuator and be able to streamline the intensity of sunlight that can be absorbed by solar cells with the identification system that can be done in realtime or directly to minimize the occurrence of power wastage. Prediction is one of the most important elements for decision-makers of the problems that occur above. In this paper, the method used in predicting the sun's angle is by using the Recurrent Neural Network (RNN) method which has a Long-Short Term Memory (LSTM) structure in the RNN system. In this reseacrh, the data used to predict the sun angle from the sunscalc.org site in the area of the University of Muhammadiyah Malang campus 3 with a period of one year. Testing with the RNN-LSTM structure is done with two different prediction models, namely weekly, monthly, and annual data for daily data results, and hourly data for daily data for one week. The test results on weekly data detect Root Mean Square Error of 0.12%, the monthly data is 0.1% and the annual data is 0.24%. The monthly data model has the fewest error values, so predictive data has a high degree of accuracy.