基于多层长短期记忆神经网络的用水量预测  

Water Consumption Prediction Based on Multi-layer Long and Short Term Memory Neural Network

作　　者：王健[1] 刘丽[1] 查淳膺 陈国炜[1] WANG Jian;LIU Li;ZHA Chun-ying;CHEN Guo-wei(School of Civil and Hydraulic Engineering,Hefei University of Technology,Hefei 230009,China)

机构地区：[1]合肥工业大学土木与水利工程学院,安徽合肥230009

出　　处：《水电能源科学》2023年第12期24-27,共4页Water Resources and Power

基　　金：安徽省自然科学基金联合基金项目(2208085US13)。

摘　　要：及时准确的居民用水量预测对供水系统的设计和运行管理至关重要。长短期记忆神经网络(LSTM)是一种有效的基于数据驱动的用水量预测模型,但其通常依赖于大量的参数设置。因此,在LSTM模型基础上,通过叠加时间分布模块,提出多层长短期记忆神经网络模型(MLSTM)。与LSTM模型对比分析表明,MLSTM模型具有较低复杂度和更高的预测精度,尤其对于高峰期用水量预测(M_(MAPE)值降低约60%),且受外部环境条件(如天气)的影响较小。Timely and accurate forecasting residential water consumption is critical to design and operational management of water supply systems.Long short-term memory(LSTM)is an effective data-driven prediction model for water consumption,but it usually relies on a large number of parameter settings.This paper proposed a multilayer long shortterm memory neural network model(MLSTM),which was built on the LSTM model by superimposing a time distribution module.The results indicate that the MLSTM model has lower complexity and higher prediction accuracy than the LSTM model,especially for the prediction of peak water consumption with M MAPE reduced by about 60%.Meanwhile,the MLSTM model is insignificantly affected by external environmental conditions(e.g.,weather).

关 键 词：居民用水量 长短期记忆神经网络 时间分布模块 多层长短期记忆神经网络 预测精度 

分 类 号：TU991.31[建筑科学—市政工程] TV213.9[水利工程—水文学及水资源]