基于奇异谱分析和改进的BP神经网络交通流预测

Traffic flow prediction based on the singular spectral analysis and the modified BPNN

作为智能交通的核心,短时交通流的预测研究在国内外蓬勃发展,传统的预测方法由于自身的缺陷,预测精度不容乐观。本文通过对智能群算法进行研究,针对交通流数据不确定性和非线性等特点,提出了一种奇异谱分析和改进萤火虫算法优化BP神经网络结合的预测模型。考虑到一些交通流数据的缺失和统计误差,利用奇异谱分析(SSA)处理数据,剔除噪声序列,将趋势序列重构进行预测。针对传统萤火虫算法在寻优过程中易进入局部最优问题,提供了一个自适应权重优化的萤火虫算法的解决方案。通过改良后的萤火虫方法(IFA)调整了BP神经网络的权值和阈值,从而形成了基于SSA-IFA-BP的组合预测模式。实验结果表明该组合模型相比于其他的预测模式具有更好的预测精度。

As the core of intelligent transportation,the research of short-time traffic flow prediction is booming at home and abroad,and the prediction accuracy of traditional prediction methods is not optimistic due to their own defects.In this paper,through the study of intelligent swarm algorithm,a prediction model combining singular spectrum analysis and improved firefly algorithm to optimize BP neural network is proposed for the characteristics of traffic flow data uncertainty and nonlinearity.Considering the absence of some traffic flow data and statistical errors,the data are processed by singular spectrum analysis(SSA)to remove the noisy series and reconstruct the trend series for prediction.To address the problem that the traditional firefly algorithm is prone to local optimum in the process of finding the best,a solution of firefly algorithm with adaptive weight optimization is provided.The weights and thresholds of the BP neural network are adjusted by the modified firefly method(IFA),which results in a combined forecasting model based on SSA-IFA-BP.The experimental results show that the combined model has better prediction accuracy compared with other prediction models.