基于自适应模糊神经推理网络的交通流量预测研究

Traffic Flow Prediction Based on Adaptive Neuro-fuzzy Inference Systems

交通流量预测的本质是对具有非周期性、非线性和随机性的交通流量数据序列根据当前和历史数据特征对未来流量态势做出合理的判断。基于模糊神经推理网络的非线性拟合能力和推理机制,研究了自适应模糊神经推理网络ANFIS在交通流量预测中的应用。设计了3种形式的一阶模糊推理网络,对采样周期分别为30 s和2 m in的非周期性交通流量进行了预测计算,与具有不同隐层单元的BP神经网络预测结果进行了比较。结果表明自适应模糊神经网络计算简单,在交通流量趋势预测方面优势明显。

Traffic flow prediction is to judge future traffic flow trends according to the current and historical data because traffic flow data are unperodical, nonlinear and stochastic. The nonlinear fitting capability and inference mechanisms based on neuro-fuzzy inference networks apply adaptive neuro-fuzzy networks to traffic flow prediction. Three forms of one-order neuro-fuzzy inference networks were designed to calculate and predict the unperiodieal traffic flow by using the data sampled for a 30-second time interval and a two-minute time interval respectively, the results of which were compared with the prediction results of the BP neural networks with various hidden layers. The comparison indicates that adaptive neuro-fuzzy inference networks need only simple calculation and have obvious advantages for traffic flow trend prediction.