Eventos Anais de eventos
COBEM 2023
27th International Congress of Mechanical Engineering
COMBINED MACHINE LEARNING AND DECOMPOSITION MODELS BY PREDICTING URBAN WATER CONSUMPTION IN CURITIBA
Submission Author:
Wilton Brayner , PR , Brazil
Co-Authors:
Wilton Brayner
Presenter: Wilton Brayner
doi://10.26678/ABCM.COBEM2023.COB2023-2036
Abstract
Forecasting water consumption is an extremely important factor for the planning of the agencies that distribute water to each neighborhood in each city. Recently, sanitation companies have invested in automating their water supply systems, which extract and store water consumption data in real-time. To make correct decisions regarding demand forecasting, it is necessary to have prior knowledge of consumption throughout the days considering seasonality such as holidays, weekends, also different needs according to the seasons of the year. According to the Institute for Research and Urban Planning in Curitiba now is formed seventy-five neighborhoods being Bairro Alto one of the most populated of the city, with forty-two thousand inhabitants, which represents 2.15% of the total population, the neighborhood studied in this research. This paper aims to present different short-term water consumption forecasting models using different forecasting techniques such as Holt-Winters, Autoregressive Integrated Moving Average (ARIMA), Random Forest, and XGBoost for different forecasting horizons, one, seven, fourteen, and twenty-eight steps ahead for daily data in one time series about water distribution for SANEPAR to Bairro Alto in Curitiba. These predictions are based on historical data collected over three consecutive years to predict urban water consumption in the previously mentioned neighborhood. The partial results obtained were 0.39 for the R2 with daily data and 0.96 for hourly data considering the ARIMA model and 0.28 for the R2 with daily data and 0.92 for hourly data considering the Holt-Winters model.
Keywords
Time series forecasting, machine learning, Decomposition models, water supply
