DESIGN A DEEP LEARNING LONG SHORT-TERM MEMORY (LSTM) FOR FLOW PREDICTION AND ANOMAL DETECTION IN WATER SUPPLY NETWORK

Abstract

In this paper, we designed an LSTM deep learning network to predict time series data which is the water flow of clean water supply networks. Based on the prediction results, a model was built to detect anomalies. The model was tested on 3 points measuring water flow of water supply network in Hue city with low predictive error and high NSE index of 0.98. Predictive results of the model were used to build anomaly detection model in the network based on predictive errors and real data. Experimental results show that the proposed method gives detection results with high precision, which can be applied in practice.