Yenilikçi dl ilkeleriyle güneş pv sistemlerinden üretilen elektrik gücünün tahmini

This master's research is implemented to examine and explore the significant contributions and substantial relevances of novel DL algorithms in making predictions with considerable effectiveness, accuracy, reliability, and trustworthiness. The work addressed a case study comprising two solar PV systems in India. Datasets were defined, and two performance evaluation approaches were used to examine the accuracy of the three DL models: LSTM, CNN, and LSTM-CNN. The numerical simulations and mathematical analysis in this work revealed that the LSTM, CNN, and hybrid LSTM-CNN models offered higher accuracy in forecasting clean electrical power produced from the two solar PV systems. Nonetheless, the accuracy of the hybrid scheme provided the most significant accuracy rates compared with the first two algorithms. The research findings also indicated that the values of MAE linked to the training of the three algorithms were more considerable than the MAE amounts related to the testing process across all the epoch ranges. Besides, it was found that the MAE connected with training for the three algorithms started at a maximum value. Then, it declines until it reaches a lower value steadily for a more extended epoch range. However, the MAE value after declination was still larger than the MAE rates of the testing process. Further, the numerical outputs confirmed that the RMSE value of the training and testing procedures for the three algorithms had similar behavior to the MAE in forecasting the clean electrical power of the two solar PV systems.