Developing an LSTM model to forecast the monthly water consumption according to the effects of the climatic factors in Yazd, Iran
Azar Niknam, Hasan Khademi Zare, Hassan Hosseininasab, Ali Mostafaeipour
Abstract
Forecasting water demand based on past consumption patterns is one of the main methods of planning the water supplies in urban areas, especially where water shortage and multiple droughts occur. Water consumption is affected by different environmental factors of which one may refer to climatic factors. In this purpose, two models were designed to predict water consumption with regard to the role of climatic factors. In the pre-processing stage, the outlier data were identified and smoothed in a box diagram, then the relationship between climatic factors and water consumption was investigated through Pearson’s correlation coefficient. In the analysis stage, two innovative hybrid methods known as Univariate-LSTM (UV-LSTM) and Multivariate-LSTM (MV-LSTM) were adopted to predict the monthly water consumption via two methods, including the Long Short-Term Memory (LSTM) network and the correlation coefficient results. The monthly water consumption by the urban users and the climatic factors in Yazd, Iran, from 2011 to 2020 were used as the data to test the proposed models. The results showed that the average temperature had the greatest positive effect on the water consumption, so it was considered as an input variable in the MV-LSTM model. Also, comparing the predictive performances of the two models via the root mean squared error showed that the MV-LSTM model outperforms the UV-LSTM.