基于集成LSTM的泵站供水流量智能预测方法

Intelligent Prediction Method of Water Supply Flow in Pump Station Based on Integrated LSTM

城市二次加压供水泵站的供水流量预测是实现清水池补水、蓄水的依据,也是保证居民用水安全的前提.针对泵站供水流量受线性、非线性和时变等多种因素影响,导致传统模型的预测效果较差的问题,提出了一种基于长短时记忆网络与整合移动平均自回归模型相结合(LSTM-ARIMA)的方法,建立泵站供水流量集成预测模型.首先将获取到的供水流量数据按照时间日期进行打标签及预处理;然后将处理后的数据分别放入LSTM模型和ARIMA模型中进行训练与测试,通过统计分析2个模型的历史预测准确次数来确定它们各自的基本权重,并在预测过程中自适应修正权重;最后,基于对应权重将2个模型集成,得到最终的供水流量预测结果.某供水泵站的现场数据验证表明:本文方法所得结果与其他2种方法所得的预测结果在均方根误差(RMSE)上分别降低了51.24%和66.52%,在平均绝对误差(MAE)上分别降低了49.84%和67.02%,验证了模型的有效性.

The prediction of water supply flow of urban secondary pressurized water supply pump station is the basis to realize water replenishment and storage of clear water pool,and also the premise to ensure water safety of residents.Aiming at the problem that the water supply flow of pumping stations is affected by linear,nonlinear and time-varying factors,which leads to the poor prediction effect of traditional models,a method based on the combination of deep learning long and short-term memory network and integrated moving average autoregressive model(LSTM-ARIMA)is proposed to establish an integrated prediction model of water supply flow of pumping stations.Firstly,the water supply flow data obtained are labeled and preprocessed according to the time and date.Then,the processed data are put into LSTM model and ARIMA model respectively for training and testing.The basic weights of the two models are determined by statistical analysis of the historical prediction accuracy times of the two models,and the weights are modified adaptively in the prediction process.Finally,the two models are integrated based on the corresponding weights to obtain the final water supply flow prediction results.The field data of a water supply pump station is used to verify the effectiveness of the proposed method.The results obtained by the proposed method are compared with the prediction results obtained by the other two methods.The results show that the RMSE is reduced by 51.24%and 66.52%respectively,and the MAE is reduced by 49.84%and 67.02%respectively,which verifies the effectiveness of the model.