Meteorological element prediction for renewable energy systems: Comprehensive comparison on deep learning algorithms with/without hyperparameters tuning

Climate change greatly affects the operation of renewable energy systems (RESs). The precision of meteorological element prediction can improve the dependability and robustness of renewable energy systems. In this paper, review and comparison of deep learning (DL) algorithms with or without hyperparameters tuning are presented for meteorological element prediction. First, meteorological element prediction systems based on deep learning and optimization algorithms are identified and summarized. Second, multiple types of data and multiple time dimension comparison experiments of various time series forecasting algorithms are conducted. Third, to improve the accuracy and rigor of the experiments, 4 types (solar radiation, temperature, pressure, and wind speed) of weather data collected from Basel are used for the prediction experiments, and comparative experiments are carried out over a one-month-ahead time horizon. Finally, the experiment and comparison results show the best model for solar prediction is the multilayer perceptron (MLP) model, but the nonlinear fitting model has the best performance in the prediction of temperature, pressure, and wind speed. This work improves the selection of meteorological element prediction methods based on DL algorithms for renewable energy systems and provides practitioners with advice for incorporating weather considerations into design processes.