基于LSTM-SD组合模型的城市电动汽车保有量中长期预测

Medium and Long Term Prediction of Urban Electric Vehicle Ownership Based on LSTM-SD Combined Model

电动汽车(electric vehicle,EV)保有量预测在应对EV快速发展带来的机遇和挑战中具有重要意义。然而现有的EV保有量预测存在数据缺乏、EV发展影响因素考虑不足等问题,导致预测结果可信度不足。针对此问题,文章分析了EV保有量技术、政策、环境以及社会等直接影响因素和经济发展水平等间接影响因素;分别采用长短时记忆神经网络(long short-term memory networks,LSTM)模型和系统动力学(system dynamics,SD)模型对EV保有量进行预测;基于此2种模型的误差,提出EV保有量LSTM-SD组合预测模型,提高了预测结果的精确度。基于某城市过去6年EV保有量实际数据进行算例仿真,对未来10年EV保有量进行预测,通过组合预测模型与LSTM预测模型、SD预测模型的对比实验分析,验证了所提模型的合理性,提高了EV保有量预测的精度。

The prediction of electric vehicles(EV)ownership is of great significance in dealing with the opportunities and challenges brought by the rapid development of EV.However,there are some problems in the existing prediction of EV ownership,such as the lack of data and insufficient consideration of factors affecting EV development,which lead to the lack of credibility of the prediction results.To solve this problem,this paper analyzes the direct and indirect factors such as EV ownership technology,policy,environment and society,as well as the level of economic development.Long short term memory networks(LSTM)model and system dynamics(SD)model are used to predict the EV ownership of electric vehicles respectively.Then,based on the errors of these two models,a LSTM-SD combined prediction model of EV ownership is proposed,which improves the accuracy of the prediction results.Finally,based on the actual data of EV ownership in a city in the past 6 years,an example simulation is carried out to predict the EV ownership in the next 10 years.Through the comparative experimental analysis of the combined prediction model,LSTM prediction model and SD prediction model,the rationality of the model proposed in this paper is verified,and the accuracy of EV ownership prediction is improved.