臺灣博碩士論文加值系統

　　要用一般傳統的數學模式來描述時變且非線性的負載變化，有其技術上的困難。透過神經网路對歷史資料的學習訓練，使其找出一組輸入與輸出之間的映射關係。而此訓練完成的网路將可用來做未來資料的預測工作，如此可省去繁雜的分析步驟。本論文便是利用神經网路可學習的優點來進行供電負載預測的工作。基本上，在此負載預測實驗是採用隨機倒傳遞學習法則來進行，包括了三種型態的負載預測：（一）未來二十四小時負載預測：（二）尖峰負載預測：（三）未來一小時負載預測。

　　It''s technically difficult to apply the classical mathematics to characterize the behavior of the power load system. By using the historical power load data to train the Network, the mapping relationship between the input and output can be found automatically. Basically the Stochastic BP learning algorithm are facilitated to go on the experiment. In this thesis we are to do three types of power load prediction, including (1) 24-hour ahead load prediction; (2) Peak load prediction;(3) One-hour ahead load prediction.