摘要
针对船舶交通流时间序列的非线性和非平稳性特点,设计一种结合集合经验模态分解(ensemble empirical mode decomposition,EEMD)和差分进化算法优化BP神经网络(back propagation neural network optimized with differential evolution algorithm,DEBPNN)的船舶交通流组合预测模型(EEMD-DEBPNN).首先,利用EEMD算法降低船舶交通流时间序列的非平稳性;然后,对EEMD分解后获得的各非线性分量采用DEBPNN模型(先采用DE算法对BPNN的初始权值和阈值进行预寻优,再利用预寻优获得的初始权值和阈值训练BP神经网络得到最优的权值和阈值)进行预测;最后,再将各分量预测值进行叠加即得到最终预测结果.基于长江某港口航道船舶月交通流数据,将该组合模型与BPNN、DEBPNN模型进行实例对比分析.结果表明,EEMD-DEBPNN较DEBPNN、BPNN模型的预测精度更高.
Aiming at the nonlinear and non-stationary characteristics of vessel traffic flow,a vessel traffic flow combination prediction（ EEMD-DEBPNN） model was designed by assembling ensemble empirical mode decomposition（ EEMD） algorithm and back propagation neural network optimized with differential evolution（ DEBPNN） algorithm for predicting vessel traffic flow more accurately. Firstly,the EEMD method was used to reduce the non-stationary of vessel traffic flow time series,and then the DEBPNN model was used to predict the nonlinear components obtained after the decomposition of EEMD（ Firstly,the DE algorithm was used to pre optimize the initial weights and thresholds of BPNN,and then the initial weights and thresholds obtained from pre optimization were used to train BP neural network to get the best weights and thresholds）,and finally,the predictions for all components were added up and the accumulation result was namely the final prediction of the EEMD-DEBPNN model. Based on the statistical data of monthly vessel traffic flow from a certain port of Yangtze River,the EEMD-DEBPNN prediction result was compared with those of BPNN and DEBPNN models,and the results show that the EEMD-DEBPNN model has higher prediction accuracy.
作者
肖进丽
李晓磊
XIAO Jin-li;LI Xiao-lei(School of Navigation, Wuhan University of Technology Wuhan 430063, Chin;Hubei Key Laboratory of Inland Shipping Technology, Wuhan 430063, Chin)
出处
《大连海事大学学报》
CAS
CSCD
北大核心
2018年第2期9-14,共6页
Journal of Dalian Maritime University
基金
湖北省自然科学基金面上项目(2015CFB282)
