基于RBF神经网络的交通流数据修复研究

完整的传感器数据是交通管理和控制的基础，但由于传感器自身或传输线路故障等原因，常常导致数据缺失，亟需对传感器缺失数据进行修复。鉴于此，以离散和连续缺失的线圈检测器交通流量数据为研究对象，构建基于RBF神经网络的数据修复模型。并将其结果与利用非线性回归模型、BP神经网络模型进行修复的结果相比较。RBF神经网络模型在离散缺失3 个数据、连续缺失3 个数据和连续缺失5 个数据情况下，平均百分比绝对误差分别为0.67%, 0.66%和1.33%，修 复值和实测值的总体相关性为0.992，修复精度优于非线性回归模型和BP神经网络模型。研究结果表明，RBF神经网络模型与其他方法相比可更精确地进行交通数据修复。

Repair of Traffic Flow Data Based on RBF Neural Network

Complete sensor data is the basis for traffic management and control. Because of the sensor itself and transmission line failures, data is often missed and needs repairing. Given this, RBF neural network model was developed to repair discrete and continuous missing data of inductance loop detector. The results of this model were compared with that of non-linear regression model and BP neural network model. It shows that when the loop detector outputs miss three discrete data, three consecutive data and five consecutive data, the percentage of the average absolute error for RBF neural network model are 0.67%, 0.66% and 1.33% respectively; correlation between repaired value and measured value is 0.992. The repair precision of RBF neural network model is superior to that of the nonlinear regression model and BP neural network model. Therefore, RBF neural network model can repair missing data more accurately compared with other methods.