基于ISSA-LSTM模型的短时交通流预测

Research on Traffic Flow Prediction Based on ISSA-LSTM Model

针对现有模型对短时交通流预测精确度不高、模型参数难以确定的问题,提出一种基于改进的麻雀搜索算法(ISSA)和LSTM的短时交通流预测模型(ISSA-LSTM)。使用改进的ISSA算法优化LSTM的关键参数,减少参数的不确定性,从而构建预测精度高的交通流预测模型。该模型具备LSTM提取时序数据深层特征的能力,融合了SSA算法快速收敛和全局搜索的特点,并且改进了SSA算法麻雀最初位置分布完全随机的特点,使其能均匀地分布在各个区间,避免出现局部最优的可能。在真实的交通流数据集上进行验证,将模型的预测结果与BP、GRU、LSTM、PSO-LSTM和SSA-LSTM网络的预测结果进行对比。实验结果表明,ISSA-LSTM模型的RMSE相较于LSTM模型下降了3.263,MAE下降了1.87,MAPE下降了0.949百分点,R^(2)上升了0.276百分点,相较于其他对比模型,组合模型的RMSE、MAE、MAPE、R^(2)的评价指标效果均为最好。因此,ISSA-LSTM模型在短时交通流预测上有较高的预测精度和预测稳定性,对预测交通流有借鉴意义。

Aiming at the problems that the existing models have low accuracy in predicting traffic flow and the model parameters are difficult to determine,a traffic flow prediction model(ISSA-LSTM)based on an improved Sparrow Search Algorithm(ISSA)and LSTM is proposed.The ISSA is used to optimize the key parameters of LSTM and reduce the uncertainty of the parameters,so as to build a traffic flow prediction model with high prediction accuracy.The model has the ability of LSTM to extract deep features of time series data,combines the characteristics of fast convergence and global search of SSA algorithm and improves the characteristics of SSA algorithm that the initial position distribution of sparrows is completely random,so that it can be evenly distributed in each interval to avoid occurrence of possible local optima.Validated on a real traffic flow dataset,and compared the prediction results of the model with the prediction results of BP,GRU,LSTM,PSO-LSTM and SSA-LSTM networks.The experimental results show that compared with the LSTM model,the RMSE of the ISSA-LSTM model decreases by 3.263,the MAE decreases by 1.87,the MAPE decreases by 0.949 percentage points and the R^(2) increases by 0.276 percentage points.The evaluation indicators of RMSE、MAE、MAPE and R^(2) of ISSA-LSTM model are both the best compared with other comparison models.Therefore,the ISSA-LSTM model has higher prediction accuracy and prediction stability in short-term traffic flow prediction,which has reference significance for predicting traffic flow.