基於熱傳導模型以精確偵測交通異常區域之方法

熱傳導模型主要應用於物理學上，計算隨著時間溫度的傳導過程。當給定一物體部位初始的溫度，隨著時間熱能在傳導時會有所不同，用以其物理特性計算該物體在不同部位的溫度。近年來有不少論文探討熱傳導模型在其他領域的應用，例如社群網路，交通系統，統計流形等。然而在由感測器資料與路網資訊建立的交通系統裡，要套用熱傳導模型，仍面臨所多挑戰。在交通系統中，感測器是以分佈各處的方式來建置，且由於位在路網上，感測到的交通資料具有路網的特性，如方向性、速限等特徵。如果直接套用等方向均質的熱傳導模型較為不精準，於是我們提出一個方法改良基於熱傳導模型的交通異常之偵測方法，針對路網的特性，給予感測器間不同的傳導速率及權重，藉以得到較接近真實流量的預測，進而偵測事件的發生地點。在進行實驗驗證時，我們主要與現有方法比較兩種實驗數據，其一是預測流量與實際流量的差異程度，其二是運用預測流量偵測發生交通異常地點的準確度。在實驗結果裡，平均而言，於第一種實驗數據內，我們在一天內的所有時段，均具有較貼近真實流量的預測，在第二種實驗數據裡，我們也均獲得較高的準確度，特別是當系統偵測週期設定為較短的時間區間之情形下。

關鍵字

交通流量估計；事件偵測；交通異常

並列摘要

The heat diffusion model is constructed based on the thermal conduction in physics, which computes the progress of how heat is diffused from objects with higher temperature to those with lower temperature over time. Given several metal objects of different temperatures in contact with each other, the heat diffusion model can be used to predict the temperature of the objects after a specified period of time. Recently, researchers have utilized the heat diffusion model in different application domains, such as social networks, traffic systems, and statistical manifolds. When applying the concept of the heat diffusion model to the traffic systems for anomaly region detection, many challenging issues need to be overcome. In a traffic system, sensors are deployed distributively on a road network. As a result, the sensor data contain useful features (e.g., driving direction and speed limit) that can be used for anomaly detection. Thus, our objective is to improve the heat diffusion model over a weighted directed graph, where each vertex represents a sensor, and each edge represents the distance between two sensors. In our experiments, we use two measurements to compare our work with one existing work, including the difference between our estimation and actual traffic flow, and the precision rate of anomaly detection. The experimental results show that our estimation of traffic flow is closer to the actual sensor records than that of the existing work, and our detection obtains a higher precision rate for all experimental parameters, particularly when the system detection interval is set to a shorter period of time.

並列關鍵字

無資料

參考文獻

[4] B. Jiang and J. Pei. Outlier detection on uncertain data: Objects, instances, and inferences. In Proceedings of the 27th International Conference on Data Engineering,

ICDE 2011, April 11-16, 2011, Hannover, Germany. IEEE Computer Society, 2011.

[10] R. Sparks. Spatially clustered outbreak detection using the ewma scan statistics with multiple sized windows.

[1] The freeway performance measurement (pems) system. http://pems.dot.ca.gov/. Accessed: 2015-04-24.