基于长短时神经网络的城市需水量预测应用  

Application of Long and Short Time Neural Network in Urban Water Demand Forecasting

作　　者：张薇薇 赵平伟 王景成[2] ZHANG Weiwei;ZHAO Pingwei;WANG Jlngcheng(Shanghai Chengtou Water (Group) Co.,Ltd.,Shanghai 200002,China;School of Electronic Information and Electrical Engineering,Shanghai Jiao Tong University,Shanghai 200240,China)

机构地区：[1]上海城投水务〈集团〉有限公司,上海200002 [2]上海交通大学电子信息与电气工程学院,上海200240

出　　处：《净水技术》2019年第A01期257-260,286,共5页Water Purification Technology

基　　金：上海市科委项目“多水源连通与原水系统智能调度技术研究与应用”(17DZ1202700)

摘　　要：在分析影响居民用水量相关性因素的基础上,采用长短时神经网络结合Encoder-Decoder方法建立城市需水量预测模型。长短时神经网络可以自动从时间序列的历史数据中抽取数据特征,避免了手动设计输入变量特征的繁琐,且可以采用更长时间的历史数据进行训练,充分考虑长期条件下不同天气、节假日的城市居民用水特征。Encoder-Decoder的网络结构模拟大脑对数据处理和做出决策的过程,适合多小时水量预测模型的构建。该模型应用于某地区需水量预测,取得了较高的预测精度,模型的适用性得到了有效验证。The high-precision urban hourly water demand forecasting model is established by using the long-short neural network combined with the Encoder-Decoder structure.The long and short time neural network can automatically extract the data features from the historical data of the time series,avoiding the trouble of manually designing the characteristics of the input variables,and can use the historical data for a longer time to train,fully considering residential water consumption characteristics under different weather and holidays.Encoder-Decoder^s network structure simulates the brain^s process of data processing and decision making,and is suitable for the construction of multi-hour water prediction models.The model is applied to the prediction of water volume in a certain area,which effectively validates the long-short neural network algorithm for urban water demand forecasting.

关 键 词：需水量预测 长短时神经网络 Encoder-Decoder 

分 类 号：TU731.5[建筑科学—建筑技术科学]