基于麻雀搜索算法优化长短期记忆网络模型的充电桩零部件采购需求预测  

Forecast of Procurement Demand for Charging Pile Parts Based on Ssa-Lstm Model Model

作　　者：王媛 李俊达 鸭磊 吴全才 吕彦颉 王慧杰 Wang Yuan;Li Junda;Nia Lei;Wu Quancai;Lyu Yanjie;Wang Huijie(Yunnan Power Grid Co.,Ltd.,Chuxiong Power Supply Bureau,Chuxiong 675000,Yunnan,China)

机构地区：[1]云南电网有限责任公司楚雄供电局,云南楚雄675000

出　　处：《云南电力技术》2025年第1期61-67,共7页Yunnan Electric Power

摘　　要：合理的充电桩零部件采购计划直接关系到充电设施的规划、建设和运营效率。准确的零部件需求预测有助于运营商优化库存,降低成本,避免维修和更新延迟。因此提出了基于改进SSA-LSTM的需求预测模型。针对LSTM超参数进行人为选取时,会造成精度过低这一问题,引入了麻雀优化算法(SSA)对神经元个数、学习率、正则化系数进行优化,它可以自动适应不同数据的特点并且具有较好的全局搜索能力和较快的收敛速度,能得到较高的预测精度;然后将训练好的模型用于零部件的需求预测。实验结果表明,该改进模型用于充电桩零部件的需求预测相较于LSTM模型有更好的预测效果,具体为MAE降低了0.04663、MSE降低了0.01394、RMSE降低了0.01264、R2升高了0.00261、MAPE值降低了2%,相比之下该改进模型更适用于充电桩零部件需求预测。提出的模型为充电桩零部件采购行业提供了更好的需求预测方案。A rational purchasing plan for charging pile components is crucially linked to the planning,construction,and operational efficiency of charging facilities.Accurate demand forecasting of components enables operators to optimize inventory,reduce costs,and avoid delays in maintenance and updates.In order to enhance the precision of demand purchasing for its components,a demand forecasting model based on enhanced SSA-LSTM is proposed.Manual selection of hyperparameters for LSTM often results in excessively low accuracy.To address this issue,the Sparrow Search Algorithm(SSA)is introduced to optimize the number of neurons,learning rate,and regularization coefficient.This algorithm can automatically adapt to different data characteristics,possesses better global search ability and faster convergence speed,and can achieve higher prediction accuracy.Subsequently,the trained model is utilized for demand forecasting of components.Experimental results demonstrate that the improved model has a superior prediction effect for the demand forecasting of charging pile components compared to the LSTM model.Specifically,it reduces MAE by 0.04663,MSE by 0.01394,RMSE by 0.01264,and increases R2 by 0.00261.The MAPE value also decreases by 2%.Therefore,the improved model is more suitable for demand forecasting of charging pile components.In conclusion,the proposed model offers a superior solution for demand forecasting in the charging pile component purchasing industry.

关 键 词：LSTM模型 麻雀搜索算法 需求预测 预测模型 

分 类 号：TM74[电气工程—电力系统及自动化]