【摘要】 针对交通流数据存在的随机性与非线性等导致短时交通流预测精度不高的问题，给出一种多阶段优化策略和改进澳洲野狗算法(Improved Dingo Optimization Algorithm, IDOA)优化LSSVM、LSTM和XGBoost参数的组合预测模型(MO-IDOA-LLX)。使用变分模态分解(Variational Modal Decomposition, VMD)将交通流分解，借助样本熵(Sample Entropy, SE)将子序列重组，得到趋势、细节和随机分量并采用相空间重构算法(Phase Space Reconstruction, PSR)对其进行处理。通过4个基准函数验证IDOA算法性能。对重构后的分量分别建立IDOA-LSSVM,IDOA-LSTM以及IDOA-XGBoost三个子模型，叠加各子模型的预测值得到预测结果。实验结果表明：其它预测模型相比，上述模型预测精度均有不同程度的提升，输出的预测结果更接近真实值。更多还原

【Abstract】 A multi-stage optimization strategy and an Improved Dingo Optimization Algorithm(IDOA) are proposed to optimize the combined prediction model with LSSVM,LSTM and XGBoost parameters, which is called MO-IDOA-LLX and is able to address the problem of low accuracy of short-time traffic flow prediction due to the randomness and nonlinearity of traffic flow data. The traffic flow is decomposed using Variational Modal Decomposition(VMD) and the subsequence is reorganized with the help of Sample Entropy(SE) to obtain the trend, detail and stochastic components and to reconstruct the traffic flow using Phase Space Reconstruction(PSR) algorithm. The performance of the IDOA algorithm is verified by four benchmark functions. Three sub-models, IDOA-LSSVM,IDOA-LSTM and IDOA-XGBoost, are built for each reconstructed component, and the prediction results are obtained by superimposing the prediction values of each sub-model. The experimental results show that compared with other prediction models, the prediction accuracy of our model has different degrees of improvement, and the output prediction results are closer to the real values.更多还原