Clean Water Demand Prediction Model Using The Long Short Term Memory (LSTM) Method

Abstract

Cities or districts as population centers with various service facilities, really need the provision of clean water. The agency that handles clean water in Indonesia is the Regional Drinking Water Company (PDAM). PDAMs were established in every city and district in Indonesia as agencies that serve the community's need for clean water. One of them is the Regional Public Company (Perumda) Tirta Pakuan and as time goes by the number of customers will definitely increase so that the need for clean water will also increase. The purpose of this research is to create a Clean Water Demand Prediction Model using the Long Short Term Memory (LSTM) Method to find the most optimal modeling. The data in this study were obtained from data reports is from Perumda Tirta Pakuan. The prediction model development process is carried out through Visual Studio Code tools. To find a model with the smallest error rate using various ratios, namely 80:20, 70:30, 60:40, and 50:50, then testing is also carried out based on the number of different hyperparameter values in batch sizes 5, 10, 15, 20, 25 and max epoch 50, 100, 150, 200, 250. From all the experiments that have been carried out, the most optimal is batch size 5 and epoch 50 with a ratio of 60:40 for water production to get RMSE 0.4862 and MAPE 2.5252% while for the amount of water use with a ratio of 50:50 get RMSE 0.4674 and MAPE of 2.5163%.