高速公路行車速率快，一旦有事件發生，往往會造成嚴重的車流延滯以及人員和財務的損失。因此若能快速偵測事件發生，將有助於交通管理者迅速採取適當措施，減低意外所造成的影響。本研究的目的為嘗試以灰色預測和假設檢定的方法建立一套事件偵測演算法，來判斷有無事件發生，並以實例測試灰色預測於事件偵測的績效。演算法的偵測邏輯為：根據最近幾個時段的佔有率進行預測，並比較該預測值與實際觀測值的差異，以兩者之間的差異程度為判斷事件發生與否的標準。整個演算法的架構包括灰色預測模式和門檻值界定兩大部分。演算的流程為建立歷史佔有率預測誤差分配圖，利用假設檢定方法算出門檻值，再將灰色預測算出來的預測值與實際值做比較之誤差值，和門檻值比較，判斷有無事件發生。本研究根據不同型Ⅰ錯誤的機率，切割門檻值，算出不同門檻值下的偵測率和誤報率，提供給決策者做參考。決策者可依照本身所期望的偵測率和可接受之誤報率，訂定上下游佔有率預測誤差的門檻值，來判斷有無事件發生。 The car on freeway is traveling at high speed. Once an incident happens, it will cause serious traffic delay and the loss of life and financial affairs. Therefore, if it can detect the occurrence of incident happen as soon as possible, it can help traffic managers to make appropriate decision to decrease accident influence. This research build up an incident detection algorithm by using grey prediction and hypothesis test, which judged whether incidents happened or not. We use actual examples to test the effect of grey prediction in incident detection. The detected logic of this algorithm is to predict occupancies according to late time and compare the difference between predict value and actual value. The standard determined to occur incident is base on the difference. The whole framework of algorithm includes grey prediction and threshold decision. The procedures in this algorithm are built up historically predictive occupancy error distribution, then using hypothesis test for threshold. Moreover, compared errors that actual value subtracts predict value of grey prediction with the threshold to judge whether anything happened. This research depends on different probability of typeⅠerror cutting threshold and figuring out detection rate and false-alarm rate for policymaker’s consulting. The policymaker makes upstream and downstream thresholds according to his expectant detection rate and acceptable false-alarm rate to judge whether incident occurred.
