基于LSTM车速预测的在线优化能量管理策略研究

Research on Online Optimization Energy Management Strategy Based on LSTM Velocity Prediction

为充分挖掘48 V混合动力汽车的节油潜力,提出了基于长短期记忆神经网络(LSTM)车速预测的混合动力汽车在线优化能量管理策略。首先,引入LSTM建立了车速预测模型,并综合考虑模型输入、输出序列长度对预测精度的影响,实现对车速的准确预测;然后,为了实现能量管理的在线最优化并提高控制策略的实时性,提出了采用快速动态规划算法的预测控制策略(FDPS);最后,分别以标准循环工况WTVC和实车工况进行测试,验证了所提策略的有效性。研究结果表明,相比于径向基神经网络(RBF)预测模型,所建立的LSTM的车速预测模型可提高29.06%-35.31%预测精度;在WTVC工况和实车工况下,所提出的FDPS策略可使燃油经济性相比于常用的规则策略(RB)分别提高5.62%和12.86%,计算时间相比于传统的预测控制策略(MPC-DP)缩短65.84%以上。

In order to fully tap the fuel-saving potential of 48 V hybrid electric vehicles, an online optimization energy management strategy for hybrid electric vehicles based on long short-term memory(LSTM) neural network velocity prediction is proposed. First, a velocity prediction model based on LSTM is established, and the influence of model input and output sequence length on prediction accuracy is comprehensively considered. Then, in order to optimize energy management online and improve the real-time performance of control strategies and realize online applications, a fast dynamic programming algorithm is proposed to solve the predictive energy management optimization problem. Finally, the standard vehicle driving cycle WTVC and the collected vehicle driving cycle are tested to verify the effectiveness of the strategy proposed in this paper. The research results show that the proposed LSTM-based velocity prediction model has an accuracy improvement of 29.06%~35.31% compared to the radial basis function(RBF) neural network prediction model. At the same time, the proposed online optimization energy management strategy has a fuel-saving rate improvement of 5.62% and 12.86%compared with the rule-based(RB) strategy under the test cycles and the calculation time is shortened more than 65.84% compared with tradition predictive control strategy(MPC-DP).