Application of Multilayer Perceptron Artificial Neural Networks to Mid-Term Water Consumption Forecasting - A Case Study

作者：Piasecki Adam; Jurasz Jakub; Marszelewski Wlodzimierz^*

来源：Ochrona Srodowiska, 2016, 38(2): 17-22.

• Water supply system
• water consumption
• forecasting
• artificial neural networks
• exogenous variable

Multilayer perceptron (MLP) artificial neural networks were employed to monthly water consumption forecasting. Research encompassed Czerniewice, one of the estates in Torun with a dedicated waterworks system (different from the other part of the town). Initially, nine exogenous variables describing meteorological, economic and social conditions were examined. The forecasting process revealed that implementation of all input variables correlating with water consumption did not lead to the highest quality forecasts. In terms of quality, the best result (evaluated based on MAPE criterion) was achieved for a model built on variables such as number of residents with access to waterworks, water rate, maximum temperature and humidity, and average income per inhabitant. It was demonstrated that the selection of input variables used for water consumption forecasting should be adjusted to local conditions. In the example considered, artificial neural networks proved useful in mid-term water consumption forecasting.

• 出版日期2016