基于CEEMDAN-ABC-LSTM组合模型的短时交通流预测

Short-term traffic flow prediction based on CEEMDAN-ABC-LSTM combined model

为了提高短时交通流预测精度,提出了基于自适应噪声完全集成经验模态分解(Complete Ensemble Empirical Mode Decomposition with Adaptive Noise,CEEMDAN)和人工蜂群算法(Artificial Bee Colony,ABC)优化长短时记忆(Long Short-Term Memory,LSTM)神经网络的短时交通流预测方法。首先将非平稳、非线性的交通流数据利用CEEMDAN算法分解成相对平稳的多个固有模态分量和趋势分量;然后用人工蜂群算法对LSTM的参数进行寻优选择,将分解后的每个模态分量分别建立CEEMDAN-ABC-LSTM模型进行预测,最后叠加每个分量的预测值输出最终的预测结果。用感应线圈实测数据对构建模型进行验证分析,实验结果表明:模型具有较高的预测性能,其平均预测精度较LSTM,ABC-SVM和ABC-BPNN模型分别提升了19.8%,25.6%和38.7%。

To improve the accuracy of short-term traffic flow prediction,a short-term traffic flow prediction method is proposed based on the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN),the Artificial Bee Colony(ABC)and the Optimized Long Short-term Memory Neural Network(LSTM).Firstly,the non-stationary and non-linear traffic flow data are decomposed into several relatively stationary natural mode components and trend components by using the CEEMDAN algorithm.Then the parameters of LSTM are optimized by using the ABC algorithm.The CEEMDAN-ABC-LSTM model is established for prediction by decomposing each modal component and optimizing the parameters of LSTM neural network.Finally,the predicted value of each component is superimposed to output the final prediction result.The experimental results show that the CEEMD-ABC-LSTM combination model has high prediction performance,and its average prediction accuracy is 19.8%,25.6%and 38.7%higher than LSTM,ABC-SVM and ABC-BPNN models respectively.