- 學位論文
基於相似天法則及類神經網路 之電價預測
A Study of electricity price forecast by applying neural network and similar-day method
摘要
中文摘要 近年來電業自由化已在世界各地展開，市場買賣行為轉變為透過競標方式的商業交易行為。在這種競標交易的模式下，區域邊際價格之預測顯得更為重要，若能精確推估區域邊際價格走勢，對市場參與者於現貨價格市場中制訂競標策略有莫大的幫助。為此，本論文旨於利用相似天迴歸模型與相似天方法結合類神經網路進行區域邊際價格之預測及比較。 本論文首先，從歷史資料中，透過四個計算相似性之數學模型，挑選與基準日相似性最高之電價和負載資料；其次，分別應用相似天迴歸模型及相似天結合類神經網路進行電價預測，以找出最佳計算相似性之模型；最後，探討相似天方法結合類神經網路預測所使用之參數、挑選相似天之區間及相似天數多寡，對於預測精確度之影響，藉以找出預測電價之最佳參數。研究結果顯示運用皮爾森相關係數計算相似性，搭配資料區間45天、相似天數3天及學習速率為0.9，以相似天方法結合類神經網路進行區域邊際價格預測時，可得到最佳預測精確度。
並列摘要
英文摘要 Recently, electric deregulation has been a great impact on electricity industry worldwide, and the electricity industry market has changed from traditional business model to a public bidding competition. This new transaction mode has been popular over countries. Because of bidding competition, the locational marginal prices (LMP) forecast is becoming more and more important. The tendency of LMP is significant for the market participants to develop their bidding strategies. This paper presents similar day regression method (SDRM) and similar day combined with artificial neural network (SANN) to forecast and compare the locational marginal prices. Firstly, the electricity price and loading data which are the most similar to the data of the basis day are selected by using four similar mathematical calculation models. Secondly, SDRM and SANN are applied to forecast electricity price in order to obtain the best similarity model. Finally, the parameter of SANN, the selected interval of similar day and the amount of similar day are studies that we apply to understand the effect of forecasting accuracy and further find the optimal forecasting accuracy of electricity price. The result has revealed that using SANN method based on Pearson correlation coefficient, the optimal forecast accuracy for LMP forecast is obtained in the conditions of the intervals of the selected 45 days, the 3 selected similar days and 0.9 training rate.