Budapest Waterworks – a water consumption forecasting system

waterworks_logo

We developed a water consumption forecasting system for Budapest Waterworks to help their resource allocation process and their long term planning for the water distribution network. The system was expected to give monthly consumption estimates for the subsequent year which  were supposed to be more accurate than the numbers obtained by the planning process in place.

In the first phase of the project we analyzed the raw data: the daily water consumption numbers broken down by geographical units (Budapest districts) and by consumer type.  After compensating for the calendar effects, we examined the effects of temperature and precipitation as well as the influence of macro-economic data and water prices and we selected the right explanatory variables for our regression models:

Short term effects:

  • Monthly average temperature [°C]
  • Peak temperature [°C]
  • Precipitation [mm]
  • Days without precipitation

Long term trends:

  • Number of inhabitants
  • Economic activity
  • Water price

Budapest Water Works forecasting systemBased on these findings, we set up a hidden component state space model and we used the Kalman-filter algorithm to fit the parameters and to obtain forecast values based on historical behavior and explanatory variables. The system was built as a web application (PHP, MySQL), with the possibility to set up different scenarios for the explanatory variables. The forecast  strongly depended on the accuracy of the future values of explanatory variables.  Using a scenario, the user can select which variables to include in the model and he can also assign different weights for  specific variables.

A comparison for the monthly consumption estimate for the examined last year showed that the system has a 1.8% forecasting accuracy, as opposed to the 3.6% accuracy of previously applied planning methods of the waterworks. The improvement was even better when we excluded the industrial consumers where the pattern of water consumption is more dependent on individual pricing deals and other economic effects.

Hungary