基于深度LSTM与遗传算法融合的短期交通流预测模型

Short-term Traffic Flow Prediction Model Base on Fusion of Depth LSTM and Genetic Algorithm

短期交通流预测是交通优化控制和智能服务的基础。由于交通流日内波动性明显,使用单层长短期记忆网络(LSTM)存在泛化能力不足的问题,堆叠多层LSTM易导致模型难以快速收敛。通过对LSTM神经网络架构的优化设计,提出一种深度LSTM与遗传算法融合的交通流预测模型GA-mLSTM。首先,采用遗传算法(Genetic Algorithm,GA)对LSTM层数、Dense层数、隐藏层神经元个数和Dense层神经元个数进行优化,确定GA-mLSTM模型的网络结构设计和最优参数;然后,基于GA-mLSTM模型的预测结果,使用差分运算对预测误差进行修正;最后,利用公开数据集对交通流预测性能进行评估并验证,实验结果表明:GA-mLSTM模型采用3层LSTM神经网络结构,融入遗传算法和差分计算后,能有效捕获路网交通流的波动特性,可以实现更准确的交通流预测。

Short-term traffic flow prediction is the basis for optimizing traffic control and intelligent services.Due to the significant intra-day volatility of traffic flow,the use of a single-layer Long Short-term Memory network(LSTM)suffers from insufficient generalization ability,and stacking too many LSTMs will make the model difficult to converge quickly.Through the optimization design of LSTM neural network architecture,this paper proposes a traffic flow prediction model GA-mLSTM based on the fusion of deep LSTM and genetic algorithm.Firstly,the number of LSTM layers,the number of Dense layers,the number of hidden layer neurons,and the number of Dense layer neurons are optimized by using the Genetic Algorithm(GA),the network structure and optimal parameters of the GA-mLSTM model were determined;then,based on the prediction results of the GA-mLSTM model,the prediction errors were corrected by using differential processing;finally,the traffic flow prediction performance was evaluated and validated by using public data sets.The experimental results show that the GA-mLSTM model adopts a three-layer LSTM neural network structure can effectively capture the fluctuation characteristics of road network traffic flow and achieve more accurate traffic flow prediction after integrating genetic algorithm and differential calculation.