Güneş ışınımı üzerinde derin öğrenme kullanarak zaman serileri tahmini

Electricity can be produced from fossil fuels, from nuclear energy, from bio-fuels or from renewable energy resources. As a matter of fact, energy suppliers and managers face the energy management problem. Concerning electricity generation based on solar radiation, it is very important to know precisely the amount of electricity available for the different sources and at different horizons: minutes, hours and days. Depending on the horizon, two main classes of methods can be used to forecast the solar radiation: statistical time series forecasting methods for short to midterm horizons and numerical weather prediction methods for medium to long-term horizons. In this thesis we focus on statistical time series forecasting methods. The aim of this study is to assess if deep learning can be suitable and competitive for solar radiation data time series forecasting. In this context, Recurrent Neural Network variations, namely Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models are used for time series forecasting on solar radiation data. With an experimental approach, the performance of single layer and two layered LSTM and GRU models are investigated. By hyper-parameter tuning optimal parameters are found in order to construct best model that fits the global solar radiation data. In further experiments, the data are divided into the seasons. Seasonal performance of constructed RNN models and other machine learning methods are compared and a hybrid model is proposed according to the experimental results. Finally, the effect of additional meteorological parameters on solar radiation forecasting is investigated. The results show that the LSTM and GRU models can be suitable and competitive for 1 hour horizon time series forecasting on the solar radiation data. Experiments showed that hybrid approach and additional meteorological parameters improve the performance of model.