- 學位論文
台灣電力需求預測及組合分析
The Taiwan Electricity Demand Forecasting and Portfolio Analysis
摘要
本研究使用多種時間序列方法進行電力需求預測，包括支持向量回歸（Support Vector Regression）、自回歸（Auto Regression）、移動平均（Moving Average）、向量自回歸（Vector Autoregression）以及隨機森林（Random Forest），以預測台灣2024年至2028年的電力需求。分析的能源來源包括可再生能源、燃煤、天然氣、蒸汽發電、水力發電和核能。研究結果顯示，SVR模型預測大多數能源來源呈現逐漸下降的趨勢，其中可再生能源從2024年的56.02降至2028年的54.43，核能從266.08降至263.65。相比之下，AR模型預測所有能源來源均呈現增長，核能從447.20上升至474.66。MA模型則預測核能大幅下降，從340.94降至211.66。而VAR模型預測核能穩定增長，從397.78升至446.54；隨機森林模型則預測核能輕微上升，從360.65增至376.50。模型評估指標包括均方根誤差（Root Mean Squared Error）和平均絕對誤差（Mean Absolute Error）。結果顯示，SVR模型在訓練數據上的表現最為準確，RMSE最低（114.49），MAE亦為最低（87.03）。然而，在測試數據上，各模型的RMSE值較高，最高達到196.70，表明模型在處理未見數據時的預測準確性有所下降。隨機森林模型的MAE最高（152.11），顯示其預測可靠性相對較低。本研究強調了能源需求預測的挑戰性以及採用多種方法捕捉不同趨勢的重要性，以支持台灣制定有效的能源政策和決策。
並列摘要
This study using several time series method for forecasting such as Support Vector Regression (SVR), Auto Regression (AR), Moving Average (MA), Vector Autoregression (VAR), and Random Forest to forecast Taiwan's electricity demand from 2024 to 2028. The energy sources such as renewable energy, coral burning, gas, steam electricity, hydroelectric power, and nuclear energy. The result show SVR predicts a gradual decline in most energy sources, with renewable energy decreasing from 56.02 in 2024 to 54.43 in 2028 and nuclear energy dropping from 266.08 to 263.65. In contrast, AR forecasts an increase in all sources, with nuclear energy rising from 447.20 to 474.66. MA projects a sharp decline in nuclear energy, from 340.94 to 211.66. VAR forecasts nuclear energy to increase steadily, from 397.78 to 446.54, while Random Forest predicts a slight rise, from 360.65 to 376.50. Evaluation metrics, including Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE), highlight SVR as the most accurate model, with the lowest RMSE (114.49) and MAE (87.03) on training data. However, RMSE values on testing data were higher, peaking at 196.70, indicating reduced accuracy when models were applied to unseen data. Random Forest had the highest MAE (152.11).This study emphasizes the challenges of energy demand forecasting and the importance of using multiple methods to capture diverse trends to help for support Taiwan’s effective energy policy and decision making.