Development of Drainage Status Prediction Model Based on Internet of Things and Long Short Term Memory Algorithm

Abstract

The capacity of drainage can overflow due to inadequate conditions and high rainfall intensity. Several incidents in Bekasi City due to poor drainage resulted in inundation of water on the roads which resulted in damaged roads and flooding in residential areas. Several previous studies have discussed the evaluation of the drainage system using the analytical method hydrology in modeling water discharge. In most cases, the minimum capacity of the drainage canal is caused by the high intensity of rain, so the research focuses on the volume of drainage and the intensity of the rain. However, based on observations and interviews with the cleaning service, it turns out that many drainage channels are in a non-optimal condition, where there is a lot of garbage and sedimentation that hinders the flow of water when it rains. This study combines hydrological analysis modeling with drainage channel conditions whose real time data is obtained by using sensors through the internet of things (IoT). IoT devices have been able to send data well in the cloud, by combining rainfall data and then predictive modeling using RNN LSTM with training model parameters used are two layers and 20 cells with each layer given a Dropout layer with a probability of 10%. In the metric evaluation, four functions are used, namely mean squared error, Mean absolute, Nash-Sutcliffe Efficiency and Coefficient of Determination. The model has been able to see the occurrence of an increase or decrease in height and discharge. However, if you look at the results of metric calculations, the predictions generated by the model are not very good.