Abstract: A forecasting model of optimal order fractional neural network is proposed. Data pre-processing, optimal order optimization, prediction algorithm process step, and performance of predict accuracy are given. More flexible and effective than the BP neural network in function approximation ability, fractional neural network analyzes the data in two ways, time and frequency. Fractional neural network?is similar?to wavelet neural network with the good partial characteristic and distinguish rate, especially for short-term data, and has stronger self adaptation ability, faster convergence rate and higher forecast accuracy. Simulated with the short-term traffic flow, compared with based on wavelet neural network and BP neural network, performance?index?is analyzed and?evaluated. The resolution shows that traffic flow dynamic forecast based on optimal order fractional neural network is realized more flexibly, rapidly and effectively.

Key words: fractional neural network, optimal order, fractional Fourier transform, traffic flow dynamic forecast

摘要： 建立基于最优阶次的分数阶神经网络的动态预测模型，给出数据预处理、最优阶次优化和预测算法流程步骤，给定模型预测精确度的性能指标。分数阶神经网络是从时频两方面分析数据，比BP神经网络具有更灵活有效的函数逼近能力；针对短时数据分析，分数阶神经网络局部性与小波神经网络一致具有多分辨力，且有更强的自适应能力、更快的收敛速度和更高的预测精度。以短时交通流量数据为例进行仿真，与基于小波神经网络和BP神经网络模型的短时交通流量预测仿真比较，分析评价性能指标，结果表明分数阶神经网络最优阶次下可实现灵活快速有效的交通流量动态预测。

关键词: 分数阶神经网络, 最优阶次, 分数阶傅里叶变换, 交通流量动态预测