太陽能發電正逐漸成為台灣再生能源的一大主力，隨著其發電占比的增加，維持電網的穩定性上受到了一大挑戰，因此為了智慧電網的發展，太陽能發電量預測成為了一項重要議題。 本論文數據集來自太陽能發電系統中監控的多個逆變器發電量資料，因此需要設計多變量時間序列預測模型。在使用傳統建模方法的情況下，可能產生過多的計算成本與難以學習數據的多變量依賴關係，因此本論文中使用深度學習模型，結合卷積神經網路與長短期記憶循環網路預測未來一天後的逐時發電量。利用卷積神經網路提取多時間序列的特徵，而長短期記憶神經網路同時預測出多變量結果。為了更準確的預測結果，我們使用數值天氣預報中的氣象資訊進行特徵選擇，並結合天氣特徵訓練模型，並進一步根據預測日降雨條件分別晴天與雨天模型，在實驗中顯示結合氣象資訊的方式能夠使進一步降低誤差。最後考慮到台灣地區南北地區的氣候狀況差異，評估包含多個不同地區的案場的實驗結果，驗證氣象資訊結合訓練能夠廣泛應用於台灣不同氣候的地區。 ;Solar power is gradually becoming the main force of renewable energy in Taiwan. As the proportion of solar power generation increases, the challenging task of maintaining the stability of the grid becomes crucial. Therefore, for the development of smart grids, solar power generation forecasts has become an important issue. The dataset in this thesis comes from the power generation data of multiple inverters monitored in a solar power system. It is desirable to design a multivariable time series to forecast the generated power of multiple inverters. Traditional modeling methods lead to relatively high computational costs and difficulties in capturing multivariate dependencies of data. Therefore, we propose a deep learning model, which combines convolutional neural network and long- short term memory recurrent network to predict the day-ahead hourly power generation. In this model, the convolutional neural network is used to extract the features of multivariate time series, and the long-short term memory network predicts the multivariate results. In order to make the results more accurate, we use the data of Numerical Weather Prediction and perform feature selection to combine the weather features to train the model. The proposed method further separates the sunny and rainy models according to the rainfall condition of forecast day. Experiments show that the combination of weather information can further reduce the errors. Finally, considering the difference in weather conditions between the northern and southern parts of Taiwan, experiments involving multiple sites in different regions were evaluated to verify that the combination of weather information can be widely used in regions with different weather conditions in Taiwan.
