Analysis of Long-Range Forecast Strategies for IoT on Urban Water Consumption Prediction Task

Abstract

With the rapid development of technology, researchers worldwide have applied the Internet of Things to effectively transmit and monitor water levels and detect anomalies in real time. The data obtained enables numerical methods to predict water consumption as well. In the presented paper, an attempt has been made to predict water consumption for various problems forward using dedicated models and a system using an iterative approach. For this purpose, neural network algorithms such as Random Forest, XGBoost, Decision Tree, and Support Vector Regression were tested and used to train the prediction models. The results presented allowed to indicate the difference between the examined methods through the Mean Absolute Percentage Error of prediction. The used set of algorithms allowed to show the problem of estimating water prediction from different points of view. Thus, determining the tested systems’ seasonality and short-term and long-term trends. This allowed to choose the two best algorithms, one that needs less computational power to work; this seems to be a better solution.