本論文探討了天氣因素結合深度學習模型（LSTM）與時間序列模型（ARIMA）在停車場需求預測中的應用與比較。隨著城市交通問題的加劇，特別是停車難題，準確預測停車需求變得尤為重要。本研究利用台北市某停車場的歷史數據和中央氣象署氣候資料服務系統（CODIS）的天氣數據，通過LSTM與ARIMA模型對停車需求進行預測。 研究結果顯示，天氣因素對停車需求有顯著影響，特別是氣溫、降水量和全天空日射量這三個關鍵因子。在模型比較中，LSTM模型在短期預測中的性能優於ARIMA模型，特別是在數據量較小且具有較大短期波動的情況下，LSTM模型能夠更有效地捕捉時間序列中的趨勢和季節性特徵。 本研究證明了將天氣因素納入停車需求預測模型的重要性，並為未來交通管理和停車場規劃提供了重要的參考依據。同時，研究結果顯示LSTM模型在預測停車需求方面具有顯著優勢，為未來在交通需求預測中採用深度學習技術提供了堅實的理論基礎和實踐依據。 ;This thesis explores the application and comparison of weather factors combined with deep learning models (LSTM) and time series models (ARIMA) in predicting parking demand. As urban traffic problems intensify, particularly parking difficulties, accurately predicting parking demand has become increasingly important. This study utilizes historical data from a parking lot in Taipei and weather data from the Central Weather Bureau′s Climate Data Service System (CODIS) to forecast parking demand using LSTM and ARIMA models. The research results indicate that weather factors have a significant impact on parking demand, especially three key factors: temperature, precipitation, and total sky radiation. In the model comparison, the LSTM model outperforms the ARIMA model in short-term predictions, particularly in cases with smaller data volumes and larger short-term fluctuations. The LSTM model is more effective in capturing trends and seasonal characteristics in time series data. This study demonstrates the importance of incorporating weather factors into parking demand prediction models and provides crucial references for future traffic management and parking lot planning. Additionally, the research results show that the LSTM model has significant advantages in predicting parking demand, offering a solid theoretical foundation and practical basis for the future adoption of deep learning technologies in traffic demand forecasting.
