- 作者: 魏健宏; 吳耿毓
- 作者服務機構: 國立成功大學交通管理科學研究所
- 中文摘要: 現今所發展之高速公路最佳化儀控率模式,皆以複雜的數學運算來求解。本研究嘗試利用人工智慧取代目前的方式,將整個高速公路系統視為一個生命體,以具有時空特性之類神經網路學習控制策略並做出反射動作,輔以自我修正專家系統判斷行為的偏誤與決定修正方向,使整個高速公路系統具有感覺、反應與試誤性學習的能力。當系統運作時,類神經網路推估儀控率並據以實施,實施後由自我修正模組評估儀控之績效,擬定修正策略,回證到類神經網路模組中,做線上回讀訓練。經由觀察匝道儀控系統運作與分析各項績效指標,發現類神經網路系統確實能夠記取歷次錯誤之經驗,藉以修正其行為,並逐漸適應環境,朝著所設定的目標日漸改善。
- 英文摘要: Artificial intelligence, instead of arithmetic methods, is used to provide the optimal freeway rampmetering control strategy. We treat the freeway as a biological system which can learn control strategies andmake suitable decisions. An expert system is developed to evaluate the decisions and to suggest adjustingdirections. First, a neural network metering rate estimating model is constructed with time-space feature forhandling dynamic freeway traffic flow. Second, a metering rate self-adjustment expert system is formed formodifying inappropriate metering rates. Then, relevant information is sent back to the neural networkmetering rate estimating model for on-line retraining. From the simulation results, it is found that the neuralnetwork system is able to adjust control strategies toward preset targets and the overall performance isgradually improved.
- 中文關鍵字: 儀控率; 類神經網路; 自我修正; 專家系統; 匝道儀控
- 英文關鍵字: pamp metering control; expert system;neural network; metering rate; self-adjustment