Short Term Solar Power Generation Prediction Based on IOT Using SMA Optimized DNN55

Short-term forecasting includes predictions for a period of one to six hours, is essential for scheduling the power generated using solar plant. To maintain a balanced and all-encompassing operation, models that enable the reliable short-term projection of solar PV generation in the future must be developed. These circumstances increase the level of uncertainty in this solar parameter forecast. Examining deep neural networks as a potential solution for the problem of anticipating electricity demand 24 h in advance is the primary objective of this study. The real time dataset was gathered from the solar farm which contains temperature, irradiance, power, using sensors. The data quality is then improved by pre-processing which contains missing values and normalization. Then elastic net and chi-square are utilized for feature selection and validation process in the pre-processed data. These specific datasets are used to assess and train the improved DNN55 classifier, which predicts solar power. The accuracy of the DNN55 short-term power forecasting is increased by using data augmentation. Using performance metrics including accuracy, precision, recall, and specificity, the efficacy of the proposed methodology is evaluated; results show 95.29, 95.36, 95.37, and 97.63% respectively. Thus, the proposed deep learning technique has been optimized to detect solar power in the short term more accurately.